Since its original formulation by Bowlby (1969/1997, 1973/1998, 1980/1998), attachment theory has primarily inspired theorizing and research within developmental psychology. The pioneering work of Ainsworth, Blehar, Waters, and Wall (1978) on individual differences in children's attachment patterns has been of great importance in this respect. The notion of individual differences in attachment has been expanded by a growing literature on adult attachment patterns, and recently attachment theory has caught the attention of clinical psychologists, who are beginning to apply it to the subject of psychotherapy with adults (Fish & Dudas, 1999; Holmes, 2001; Rutter, 1995; Sable, 1992; Shane & Shane, 2001; Slade, 1999; Sperling & Lyons, 1994; Szajnberg & Crittenden, 1997; West & Keller, 1994). Consequently, the concept of adult attachment patterns has found increasing use in empirical studies of psychotherapy process as well as psychotherapy outcome, spanning different client populations and different forms of psychotherapeutic treatment. This article reviews the relevance of adult attachment patterns to individual psychotherapy with adults and the empirical findings in the field.

Although attention has mostly focused on the early relationship between children and their parents, attachment theory is a theory of the lifespan development of close relationships. In contrast to the idea that childhood dependence is ideally replaced by emotional independence in the young adult, Bowlby (1988) asserted that healthy human beings continue to rely on attachment relationships in times of danger, vulnerability, or illness. Attachment is defined as an affectional bond to another person, who is irreplaceable by others although there may be more than one such relationship. Attachment relationships are characterized by a need to maintain proximity, distress upon separation, joy upon reunion, and grief at loss. However, what especially characterizes an attachment relationship in comparison to other close relationships is the use of the attachment figure as a secure base from which to explore the world and as a safe haven to flee to in times of distress (Ainsworth, 1989; Bowlby, 1969/1997).

The tendency to form attachment relationships is thought to be biologically "wired in" because of the survival value it represented in human evolution. A hypothesized attachment behavioural system causes children to become attached to caregivers even if the caregiver does not provide the security and comfort sought for. If the caregiver is unreliable or perhaps even maltreating, the child will adjust and modify its attachment behaviour in order to obtain whatever approximation to security is possible in that particular relationship (Bowlby, 1988; Main, 1995). Attachment figures in infancy are normally parents or parent-like figures, but as development progresses, the most important attachment figures become romantic partners or close friends, and in old age sometimes one's own children (Ainsworth, 1989).

The concept of distinctive patterns of attachment initially arose out of the observational studies by Ainsworth et al. (1978) of infant—mother interaction. In an experimental separation and reunion procedure called the Strange Situation, Ainsworth identified three main patterns in children's ability to use the parent as a secure base for exploration. Children labelled B or secure comfortably use the parent as a secure base when exploring the environment. They miss their parent upon separation, seek contact with the parent upon reunion, and can be comforted by the parent and return to exploration. Children labelled A or insecure-avoidant explore the environment without referring back to the parent. They ignore the parent leaving and returning and are not visibly upset; instead, they turn their attention to toys. Children labelled C or insecure-ambivalent are focused on the parent and seem unable to explore. They are very distressed when the parent leaves, but cannot be comforted by the parent upon reunion, instead they appear angry or passive. Later a fourth category D or insecure-disorganized has been added, but children assigned to this category do not show a similar degree of organized response and are always assigned to one of the three ABC categories as well (Lyons Ruth & Jacobvitz, 1999). For detailed reviews of attachment patterns in infants and their correlates, see e.g. Greenberg (1999), Solomon and George (1999), and Thompson (1999).

Attachment patterns in infancy arise out of parent—child interaction and are regarded as a property of the specific attachment relationship—thus attachment patterns with mother and father can differ (Steele, Steele, & Fonagy, 1996). However, Bowlby believed that attachment patterns increasingly become a property of the individual rather than the relationship. This is due to the gradual formation of mental representations of attachment-related interactions—internal working models—which guide the individual in future attachment interactions (Bowlby, 1973/1998). In agreement with Bowlby's assertion, most theorists of adult attachment claim that attachment patterns in adulthood consist of generalized thoughts, feelings and expectations regulating the way that a given individual engages in close relationships (Rholes & Simpson, 2004). Adult attachment patterns are thought to be relatively stable, because new experience is assimilated to the existing working model, and because the patterns give rise to self-perpetuating interactional behaviours. Thus, a person who has experienced warm and sensitive caregiving will have an open and positive attitude towards close relationships in the future, and this will tend to elicit a positive response from others, thus confirming the positively coloured working model. Conversely, a person who has experienced repeated rejection will look for cues of further rejection, and the guarded behaviour resulting from this expectation can actually serve to elicit the feared rejection and confirm the negatively coloured working model.

The concept of attachment patterns is closely linked to the related measurement methods. In the field of infant attachment, the Strange Situation has been paradigmatic, but in the field of adult attachment, two influential strands of inquiry and measurement have developed in parallel since the 1980's. These two traditions use some of the same concepts, but in different ways and within different research fields, and readers who are not well acquainted with the sometimes subtle differences between operationalizations of adult attachment patterns might easily risk misinterpreting research findings in the area. To enable the reader to better judge the research findings on adult attachment patterns in individual psychotherapy, the different measures of adult attachment that have been applied in psychotherapy research are described in some detail here. For more extensive reviews of the existing measures of adult attachment see e.g. Crowell, Fraley, and Shaver (1999), Crowell and Treboux (1995), Hesse (1999), and Stein, Jacobs, Ferguson, Allen, and Fonagy (1998).

The Adult Attachment Interview and related methods

The first measure of attachment patterns in adulthood was the Adult Attachment Interview (AAI) devised by George, Kaplan, and Main (1996). The AAI is an hour-long semistructured interview about childhood attachment relationships and their effects. In a study of child—mother attachment, Mary Main and colleagues administered this interview to the mothers and noted some striking correspondences between a mother's discourse pattern in the interview and the pattern of attachment shown by her infant in the Strange Situation. The discourse characteristics distinguishing between mothers of infants with different attachment patterns were crystallized into a coding and classification system for the AAI (Main & Goldwyn, 1984; Main, Goldwyn, & Hesse, 2002). The coding system is thus first and foremost constructed to predict children's classification in the Strange Situation, which it does to a remarkable degree (van Ijzendoorn, 1995).

The AAI assesses states of mind with respect to attachment, which are assumed to operate partially outside conscious awareness. It yields three main classifications: autonomous, corresponding to secure attachment in infants, dismissing, corresponding to insecure-avoidant attachment in infants, and preoccupied, corresponding to insecure-ambivalent attachment in infants. Furthermore, interviews can be categorized as unresolved with respect to loss or trauma, and this classification predicts insecure-disorganized attachment in the infant (Hesse, 1999). An interview classified as unresolved is always secondarily assigned to the best fitting of the other categories. In a review of existing studies using the AAI, van Ijzendoorn and Bakermans-Kranenburg (1996) report the distribution of these categories in nonclinical low-risk populations to be 58% autonomous, 24% dismissing, and 18% preoccupied. About 19% are also unresolved with respect to loss or trauma. When the AAI has been used with clinical samples, a greater number of interviews have been difficult to classify within the original four categories. Consequently, a fifth category, cannot classify, has been added for interviews that display high levels of both dismissing and preoccupied speech or appear globally incoherent (Hesse, 1996).

The AAI coding system is notable for relying chiefly on the manner in which a person speaks about his childhood rather than the content of what is said (Hesse, 1999). The autonomous interview is characterized by coherence, collaboration with the interview process, and balance in describing favourable as well as unfavourable circumstances. The dismissing interview is marked by internal contradictions between general positive descriptions and lack of concrete episodes supporting these, an insistence upon inability to remember childhood events, and a general attempt to minimize the topic of attachment. The preoccupied interview is characterized by being long and rambling, as if the speaker is carried away by the discussion of attachment relationships. The speaker often confuses persons or slips between discussions of past and present, and lapses into extensive expressions of current anger at parents or vague and childish speech. Interviews are classified as unresolved if the person speaks in odd ways or exhibits magical thinking during the discussion of loss or trauma. Since classification relies on form rather than content, it is possible for people with traumatic childhoods to be classified as autonomous if they tell their story in a coherent and collaborative way. This in turn predicts that their children will be securely attached to them (Phelps, Belsky, & Crnic, 1998).

Based on the original coding system, Kobak, Cole, Ferenz-Gillies, and Fleming (1993) have developed an alternative coding system for the AAI in the form of a 100-item Q-sort. The Q-sort yields continuous scores on two orthogonal attachment-related dimensions: secure—anxious and deactivation—hyperactivation. The secure-anxious dimension mainly taps into the degree of coherence exhibited in the interview, whereas the deactivation—hyperactivation dimension distinguishes between the dismissing tendency to minimize attachment-related thoughts and feelings and the preoccupied tendency to be carried away by discussion of attachment relationships and show signs of current enmeshment. The Q-sort can also be used to generate the three categories autonomous, dismissing, and preoccupied, in which case it shows roughly 80% correspondence to classifications using the original system (Kobak et al., 1993).

A more recent development in the tradition of the AAI is the development of a Reflective Functioning Scale by Fonagy, Target, Steele, and Steele (1998) for application to AAI's. Reflective functioning is the ability of an individual to think of self and others in terms of mental states, and is as such not a measure of adult attachment patterns. However, it is relevant in this context since reflective functioning is closely related to security of attachment, with individuals classified as autonomous having the highest reflective functioning. High reflective functioning in the parent is actually a stronger predictor of secure attachment in the offspring than the AAI classification autonomous (Fonagy & Target, 1997).

Self-report instruments and romantic attachment styles

Coming from a background of social rather than developmental psychology, Hazan and Shaver (1987) were the first to construct a self-report instrument for measurement of adult attachment patterns in the context of romantic relationships. Their measure of adult attachment is based on the notion that individual differences in the ways adults engage in romantic relationships echo Ainsworth's ABC classifications. The original Hazan and Shaver measure of romantic attachment is a brief, forced-choice, multisentence description of three attachment styles. The secure style is characterized by comfort with closeness and dependence and absence of fear of abandonment. The avoidant style is characterized by discomfort with closeness and dependence and a feeling that others want to be "too close". The ambivalent style is characterized by fear of abandonment and wanting to be closer to others than they would like. The Hazan and Shaver self-report contrasts with the AAI tradition in assuming that attachment patterns are most relevantly measured with reference to current attachment relationships in the form of pair bonds and that individual differences in attachment are relatively accessible to consciousness and can be reliably reported in a questionnaire.

The Hazan and Shaver (1987) measure has been criticized on psychometric grounds and is recognized by most researchers, including the authors themselves, as being too simple. In response to this, multi-item Likert-type response scales were developed to measure the original construct in a more psychometrically detailed way. Collins and Read (1990) devised the Adult Attachment Scales (AAS), an 18-item questionnaire, which when subjected to factor analysis yielded three dimensions: comfort with closeness (Close), ability to depend on other (Depend), and fear of abandonment (Anxiety). Simpson (1990) and Simpson, Rholes, and Nelligan (1992) developed a 13-item questionnaire, which in a factor analysis yielded two independent attachment-related dimensions: Avoidance and Anxiety. Scores on these dimensions can be converted into prototypes corresponding to Hazan and Shaver's three styles.

Comparing the AAI with the Hazan and Shaver construct, Bartholomew and Horowitz (1991) argued that persons identified as dismissing on the AAI and persons identified as avoidant on the Hazan and Shaver questionnaire and related instruments were different in important respects. They made a distinction between dismissing avoidance, where the need for intimacy is simply denied, and fearful avoidance, where intimacy is longed for but avoided due to anxiety. They conceptualized four different attachment styles in terms of their model of self as worthy or unworthy of receiving love and protection, and model of other as willing or unwilling to supply love and protection. Secure persons have a positive model of both self and other, preoccupied persons have a negative model of self and a positive model of other, dismissing persons have a positive model of self and a negative model of other, and fearful persons have a negative model of both self and other. To measure this construct, Bartholomew and Horowitz (1991) developed the Relationship Questionnaire (RQ), which is similar to the Hazan and Shaver measure except for operating with four prototypes instead of three. The RQ was later expanded to a 30-item questionnaire, the Relationship Scales Questionnaire (RSQ) (Griffin & Bartholomew, 1994), which can be scored categorically as well as dimensionally. Bartholomew also developed a semistructured interview similar to the AAI and a related set of rating scales for assessing degree of fit with each of the four attachment prototypes. In these rating scales, content plays a greater role than in the AAI (Bartholomew & Horowitz, 1991).

Most recently, researchers working with self-report measures of adult attachment have converged on conceptualizing adult attachment in terms of two orthogonal dimensions: Avoidance and Anxiety (Brennan, Clark, & Shaver, 1998). Combinations of high and low scores on these two dimensions yield four different prototypes similar to Bartholomew's types: secure persons are low on both avoidance and anxiety, preoccupied persons are high on anxiety and low on avoidance, dismissing persons are high on avoidance and low on anxiety, and fearful persons are high on both anxiety and avoidance. However, there is increasing consensus that attachment patterns as measured by self-report methods are more appropriately conceptualized in dimensional rather than categorical terms (Fraley & Waller, 1998), so the use of distinct categories is waning in the empirical literature.

Concluding remarks

The different measures of adult attachment described here are generally well validated and reliable, but they clearly assess somewhat different constructs. Therefore, it should come as no surprise that they only correlate partially with each other (Crowell et al., 1999; Shaver, Belsky, & Brennan, 2000; Stein et al., 2002). There has been some heated debate in the field as to what is the "right" way to measure adult attachment (see Steele, 2002), but since both the interview and self-report traditions have yielded interesting findings in agreement with ideas contained in attachment theory, there seems to be merit in both approaches. Therefore, the most interesting and potentially enlightening question is not which type of measure is right, but how they are related, and what factors explain concordance and nonconcordance of classification on the two types of measures. Answers to these questions would probably help to resolve some of the ongoing discussions in the field of adult attachment, such as whether adult attachment patterns should be conceptualized in categorical or continuous terms and whether adult attachment patterns should be regarded as general trait-like styles, as more relationship-specific patterns or perhaps as a context-specific phenomenon (Stein et al., 1998).

Finally, it must be noted that even though most of the adult attachment measures lean on the Strange Situation classifications, there is only limited and somewhat contradictory evidence that persons classified in adulthood were classified similarly in childhood (Waters, Hamilton, & Weinfield, 2000; Weinfield, Whaley, & Egeland, 2004). Thus adult attachment classification cannot be taken as a reliable measure of the person's childhood relationship to parents. All measures of attachment patterns in adulthood rely in one way or another on the person's current mental representations of attachment relationships rather than on relationship behaviours like those assessed in infancy. Much still remains to be understood about the process by which infant attachment patterns are transformed into representational systems (Bretherton & Munholland, 1999).

An important impetus for adopting the concept of adult attachment into clinical psychology is the growing number of studies showing a connection between insecure attachment patterns, psychological problems, and psychopathology. Such a connection is predicted by Bowlby's (1973/1998, 1980/1998) writings, which deal extensively with the negative effects of loss or separation from attachment figures in childhood. Bowlby (1988) also described how insecure attachment relationships can lead to defensive exclusion of attachment-related thoughts and feelings and thus make the person more vulnerable to future psychological problems. Insecure attachment patterns can be conceptualized as patterns of information processing and affect regulation that developed to protect the person from attachment-related pain, but may stand in the way of happiness and adjustment in future relationships.

Attachment theory does not consider insecure attachment patterns to be psychopathological. After all, even in low-risk populations roughly half of all children and adults are assigned to one of the insecure attachment patterns (van Ijzendoorn & Bakermans Kranenburg, 1996). Furthermore, insecure attachment patterns are considered realistic adaptations to the caregiving environments in which they arise (Crittenden, 1999). It is only when the strategies inherent in the insecure attachment patterns are later rigidly and inappropriately applied to new contexts that they can be regarded as maladaptive.

Following the development of measures of attachment patterns in adulthood, several studies have explored the relation between adult attachment patterns and a range of psychological disorders and problems. On the most general level the conclusion has been that very few persons with psychological disorders are classified as secure, but that there are few clear-cut connections between diagnostic categories and specific insecure attachment patterns (Crowell et al., 1999, p. 456; Dozier, Stovall, & Albus, 1999). Nearly all existing studies of the relation between adult attachment patterns and psychopathology are correlational, so one must refrain from drawing any conclusions with respect to cause and effect in this relationship until more longitudinal studies are available. Also, since most people with insecure attachment patterns do not develop psychological disorders, insecure attachment patterns are more likely to function as general risk factors rather than sole or specific etiological factors. On the other hand, the fact that most diagnostic groups are heterogeneous with respect to attachment patterns, makes attachment pattern a potentially interesting variable in terms of matching or tailoring general treatment strategies to individual client characteristics (Meyer & Pilkonis, 2002).

Bowlby clearly meant for attachment theory to be of use in clinical work, and as a practicing psychiatrist, he himself applied it to the subject of individual psychotherapy with adults.

In A Secure Base, Bowlby (1988, p. 140) noted that: "In providing his patient with a secure base from which to explore and express his thoughts and feelings the therapist's role is analogous to that of a mother who provides her child with a secure base from which to explore the world". Since security is believed to be a prerequisite for successful exploration, attachment theory suggests that therapists should strive to build a secure attachment relationship with their clients. However, one must expect that psychotherapy clients, many of whom have a history of difficult relationships, approach psychotherapy with internal working models reflecting these difficulties. Thus, the therapist will have to be aware of and address the beliefs and expectations stemming from prior insecure relationships that stand in the way of establishing a secure working relationship. Whatever the goals of therapy are, part of the therapeutic task will usually consist in changing or supplementing the client's existing working models.

Although couched in a different terminology, the above-mentioned therapeutic implications of attachment theory are not very different from ideas found in many schools of individual psychotherapy, psychodynamically as well as cognitively oriented. Attachment theory is more insistent upon the realistic background of clients' "dysfunctional" working models than most psychodynamic theories and entails a greater explicit emphasis on the therapist—client relationship than most cognitive theories, but generally, it does not in itself suggest radically different forms of therapeutic intervention. The most original contribution of attachment theory to the practice of individual psychotherapy probably lies in the framework it provides for understanding the developmental history and internal dynamics of different forms of relationship difficulties or patterns of attachment. An understanding of attachment theory and research can sensitise the clinician to attachment themes, which are bound to be central when working with relationship difficulties (Slade, 1999).

The attachment patterns of clients can be expected to colour their help-seeking behaviour in therapy, and thus a knowledge of the dynamics of the insecure attachment patterns will help clinicians better understand their clients and might also function as a guide with respect to the choice of intervention strategies.

Furthermore, as demonstrated by the correspondence between AAI classification of parents and Strange Situation classification of their children, adult attachment patterns can be predictive of a person's ability to function as a secure base for another person who feels threatened or vulnerable. Thus, attachment patterns are also relevant to help-giving behaviour, making the therapist's own attachment pattern an important variable in the therapeutic relationship. This theoretical relevance of adult attachment patterns to individual psychotherapy forms the backdrop of an increasing number of empirical studies, which will be reviewed in the remaining part of this article.

The majority of empirical studies on the importance of attachment patterns in individual psychotherapy with adults have dealt with the ways in which client attachment patterns affect the therapeutic process. It has been hypothesized that clients with different attachment patterns will behave differently and will make different use of psychotherapy. This in turn might have predictable effects on the quality of the therapeutic alliance and on how therapists react to clients. Although theoretically just as important, fewer studies have looked into the effect of therapist attachment patterns on the therapeutic process.

Client attachment patterns and client in-treatment behaviour

The most immediate effects of client attachment patterns that have been studied in a psychotherapy process context are differences in client behaviour in therapy. Dozier (1990) studied 42 psychiatric patients' involvement in treatment, and Korfmacher, Adam, Ogawa, and Egeland (1997) studied the way that 55 mothers at-risk made use of a preventive intervention programme. In both cases, attachment patterns of clients were measured with the AAI, but in the work of Dozier the interview was scored using Kobak's Q-sort method. These studies yielded several findings corresponding to theoretical predictions.

Secure clients were judged to be more compliant with treatment than other groups (Dozier, 1990). They were more emotionally committed to treatment and accepted more help with their problems (Korfmacher et al., 1997). When comparing the more deactivating (i.e. dismissing) to the more hyperactivating (i.e. preoccupied) clients, Dozier found that higher scores on deactivation predicted clients to be less likely to seek out help and more likely to reject it, when it was offered. More deactivating clients were also generally less likely to self-disclose, and were judged by therapists to be poorer at making use of treatment. Similarly, Korfmacher et al. found that dismissing clients were less emotionally committed to treatment than secure clients were and favoured simple companionship over therapeutic interventions. Finally, Korfmacher et al. found that unresolved clients were less emotionally committed to treatment than secure clients were and were more likely than other groups to require crisis intervention during treatment.

The above-mentioned studies lend some support to the notion that attachment patterns of clients are reflected in their behaviour in treatment. Secure clients appear to be better at making use of treatment than insecure clients. Dismissing clients seem to show most difficulty engaging in a treatment relationship, whereas preoccupied clients engage emotionally but are less compliant than secure clients are. However, the empirical evidence of systematic differences in in-treatment behaviour is still quite limited. In the increasingly rich literature with theoretical descriptions and case stories to illustrate how clients with different attachment patterns behave in therapy it is often assumed that the traits characterizing attachment patterns on measuring instruments are similarly evident in a therapeutic context. But just as the home behaviour of infants can deviate importantly from their behaviour in the Strange Situation (Ainsworth et al., 1978), there might also be differences between for instance communication style on the AAI and communication in therapy. Thus, more research is needed on how clients with different attachment patterns act and present themselves in therapy.

Client attachment patterns and the therapeutic alliance

A frequent topic in the empirical literature is the effect of client attachment patterns on the formation of the therapeutic alliance. If clients with different attachment patterns approach interpersonal relationships differently, their attachment patterns can be expected to colour the therapeutic alliances they form with therapists, both in terms of quality and development over time. Another reason for inquiring into the effect of client attachment patterns on alliance formation is the empirically established relation between alliance and ultimate treatment outcome (Horvath & Bedi, 2002). Thus, possible connections between client attachment patterns and alliance might also point to outcome differences between clients with different attachment patterns.

Some of the first studies on alliance and attachment patterns used the AAS, which yields measures on comfort with intimacy, ability to depend on others, and fear of abandonment. In a sample of 76 clients at university and community counselling centres, Mallinckrodt, Coble, and Gantt (1995) found that comfort with intimacy correlated positively with alliance, whereas fear of abandonment correlated negatively with alliance. Also investigating clients from university counselling centres, Kivlighan, Patton, and Foote (1998) found that comfort with intimacy and ability to depend on others predict better therapeutic alliances (n=40), whereas Satterfield and Lyddon (1995) only found a significant relation between ability to depend on others and a positive alliance (n=60). Comfort with intimacy and ability to depend on others correlate moderately with each other and are both related to secure attachment (Collins & Read, 1990). Thus, these studies point to the overall conclusion that clients with secure attachment patterns form more positive therapeutic alliances than clients with insecure attachment patterns do, which is what one would expect based on theory. However, because of the use of the AAS, the results are not directly translatable to the more widespread categorical or two-dimensional attachment measures.

Whereas the above-mentioned studies used single point ratings of therapeutic alliance, several later studies have also inquired into changes in alliance over time. A study by Kanninen, Salo, and Punamäki (2000) investigated the effects of client attachment patterns upon the development of the therapeutic alliance in trauma therapy with 36 Palestinian political ex-prisoners. They used a paper and pencil version of the AAI to divide the clients into the clusters autonomous, dismissing, and preoccupied. Since this way of measuring attachment patterns has not been employed in other studies or compared with the AAI proper, care must be taken in interpreting the results, which were, however, quite interesting.

Kanninen et al. (2000) found no differences between the three groups on the initial ratings of working alliance, but when studying changes in alliance ratings from the beginning to the middle and end of therapy, group differences emerged. Working alliance ratings often show a high—low—high pattern (Kivlighan & Shaughnessy, 2000), and this was true of both the secure and the preoccupied group. However, both the fall in alliance towards the middle and the rise in alliance towards the end of therapy were steeper in the preoccupied group than in the secure group. The dismissing group showed a different pattern of alliance development, in which alliance was stable from the beginning to the middle of therapy, but decreased towards the end of therapy. The authors discuss the results in terms of a tendency in preoccupied individuals to show a strong emotional involvement in treatment, being extreme in negative as well as positive reactions. They speculate that the drop in alliance shown by the dismissing clients towards the end of therapy might reflect a dismissal of the importance of the therapeutic relationship in view of the coming separation.

Eames and Roth (2000) studied the relation between the attachment patterns of 30 clinical psychology outpatients and ratings of therapeutic alliance and alliance ruptures at four points in the initial phase of therapy. They found that fearful attachment on the RSQ was negatively related to working alliance, whereas secure attachment was positively related to working alliance. However, these associations only reached the level of significance for some sessions and some parts of the alliance ratings. Both dismissing and preoccupied attachment were associated with an increase in alliance over time, but the authors themselves express doubts about the reliability of these results. The clearest result of the study emerged with respect to alliance ruptures. Therapists reported significantly more ruptures with preoccupied clients and significantly fewer ruptures with dismissing clients.

In the most recent study on attachment patterns and therapeutic alliance Sauer, Lopez, and Gormley (2003) used hierarchical linear modelling to investigate the contribution of client and therapist attachment patterns to the quality of the working alliance measured at three points during the initial phase of psychotherapy at university and community counselling centres. Attachment patterns were assessed with Simpson's Adult Attachment Inventory. Sauer et al. found no effect of client avoidance or anxiety upon working alliance as rated by either the client or the therapist (effects of therapist attachment are reported below). Out of the 28 clients, 11 clients terminated therapy before the third sampling of working alliance, making the sample rather small, but the dropout clients did not differ from the completers in terms of attachment patterns or initial rating of working alliance.

With the Sauer et al. (2003) study as a notable exception, most of the remaining studies point to the expected conclusion that secure attachment in the client is related to more positive therapeutic alliances than insecure attachment. The studies further indicate that it might be feasible to study differences between the attachment patterns in terms of development and fluctuations in alliance over time, rather than in terms of levels of alliance at single point measurements. For instance, both the study by Kanninen et al. (2000) and the study by Eames and Roth (2000) suggest that alliance might fluctuate more strongly with preoccupied clients than with other client groups, although such parallels remain tentative given the large differences between the studies. Thus therapeutic work with preoccupied clients might require much more rupture repair work than therapeutic work with other clients (Safran, Muran, Samstag, & Stevens, 2002).

Eames and Roth (2000) raise the question of differential reliability of reports of therapeutic alliance by clients with different attachment patterns. For instance, secure clients might generally be reliable in their portrayal of the quality of the alliance. In contrast, dismissing clients might report good alliances in denial of their difficulty and unwillingness to engage emotionally in treatment, in a manner similar to the idealization of parents shown by dismissing persons on the AAI. Such differences would obviously complicate matters and stress the importance of supplementing client alliance rating with ratings by observers and therapists. If such a tendency to hide beneath a surface of seeming cooperation is indeed characteristic of dismissing clients, therapists would obviously do well to attend to this possibility. Future studies could relevantly inquire into whether alliance ratings by the different attachment groups differ in terms of discrepancy from ratings from the perspectives of therapists or observers.

The effect of client attachment patterns on therapist behaviour

If clients with different attachment patterns engage differently in psychotherapeutic treatment, one might also expect client attachment patterns to call forth different behaviour on part of the therapist. Whereas attachment theory generally predicts that people elicit responses from others that confirm their working models of attachment, the results might be different with therapists who have been trained not to confirm the client's worst expectations and might instead respond in an emotionally corrective way. To date two studies have looked into therapist behaviour in response to client attachment patterns. In a study by Hardy et al. (1999), client attachment patterns were assessed by analyzing client in-session discourse in a manner similar to the analysis of the AAI, and therapist responses to relationship episodes related by clients were noted. In a study by Rubino, Barker, Roth, and Fearon (2000), videotapes of actors presenting a therapeutic rupture episode as it might be experienced with each of the four Bartholomew attachment styles were shown to 77 therapists-in-training and their suggested response was noted.

Hardy et al. (1999) distinguished between secure, dismissing and preoccupied client communications styles on 10 episodes from psychodynamic-interpersonal therapy chosen by clients as being particularly helpful. They found that therapists were more likely to respond with interpretation to dismissing attachment styles and with reflection of feelings to preoccupied attachment styles. This finding confirmed an earlier finding by this group that therapists tended to react with more cognitive interventions to an underinvolved client interpersonal stance, and with more affective interventions to an overinvolved client interpersonal stance (Hardy, Stiles, Barkham, & Startup, 1998). It should be noted that Hardy et al. (1999) only evaluated attachment patterns in brief episodes taken from therapy and that client overall attachment patterns were not assessed. However, this could be quite relevant as therapists might react more directly to client attachment patterns as displayed in particular moments of therapy rather than to general attachment patterns. However, since the episodes were not chosen randomly, but based on client assessment of helpfulness, the response patterns found by Hardy et al. might not be representative of the therapeutic process in general.

Rubino et al. (2000) assessed the degree of empathy and depth of intervention shown by the therapists-in-training in their suggested response to the simulated rupture episodes. Intervention depth was defined as the degree to which the therapist goes beyond and elaborates on the information given by the client. They found that therapists responded with more empathy and with greater depth to the fearful and preoccupied than to the secure and dismissing clients. In spite of the adoption of a very different research strategy, these results show some similarities to the results of Hardy et al. (1999). Thus, the greater use of expressions of empathy with the fearful and preoccupied clients—that is, the clients hypothesized to be highest on attachment-related anxiety—might correspond to the greater use of reflection of feelings with preoccupied attachment styles in the Hardy et al. study. However, the results are more dissimilar with respect to depth of intervention. Rubino et al. found fearful and preoccupied clients to be the targets of the deepest interventions by the therapists, whereas the interpretations received by the dismissing clients in the Hardy et al. study would normally be considered deeper than the reflection of feelings received by the preoccupied clients.

Both research groups discuss their results in terms of appropriate therapist responsiveness or matching of the interpersonal style of the client. To the extent that they do find significant effects of client attachment patterns on therapist behaviour, they seem to be mostly "in style" in that therapists are more feeling-focused and empathic with anxious or preoccupied clients and are more cognitive or distant with avoidant or dismissing clients. Theoretically one might argue in favour of therapist responses "in style" to facilitate rapport, but also in favour of therapist responses "out of style" to correct or challenge the client's insecure attachment pattern, and perhaps in favour of varying response modes between different phases of therapy (Dolan, Arnkoff, & Glass, 1993). Future studies could profitably take such timing of responses in different phases of therapy into account. Furthermore, therapist responses are obviously never simple products of client attachment patterns, but also reflect the theoretical stance and personality of the therapist —with one relevant factor being the therapist's own attachment pattern.

Just like the attachment patterns of parents predict their ability to function as a secure base for their children, the attachment patterns of therapists would be expected to predict their ability to establish a secure working relationship with clients. In theory, secure attachment in the therapist is preferable, and one would expect a greater frequency of secure attachment in therapists than in client populations. Partial support for these assumptions is found in an investigation of the attachment patterns of 196 British clinical psychologists assessed with the Hazan and Shaver questionnaire, which found 69.9% to be secure, and found insecure attachment in the therapist to be related to more professional problems (Leiper & Casares, 2000). However, even in therapists classified as secure there will always be variation in degrees of attachment anxiety and attachment avoidance, and such differences in style might also affect the therapeutic process.

Similar to the relationship between client attachment patterns and therapeutic alliance, one might reasonably expect to find a relationship between therapist attachment patterns and alliance. This was investigated by Dunkle and Friedlander (1996), who studied the relation between therapeutic alliance and therapist attachment patterns as measured by the AAS in 73 therapists working in a university setting. They found that comfort with closeness in the therapist was positively related to client ratings of the emotional bond to their therapists. In contrast to this, the previously mentioned study by Sauer et al. (2003) somewhat surprisingly showed that therapist attachment anxiety as measured by Simpson's Adult Attachment Inventory was positively related to client-rated alliance in early therapy. Sauer et al. suggest that the initial good rating of alliance with therapists rated higher on attachment anxiety might be a result of the anxious therapists working harder to make the client feel good about the relationship. However, this is not productive in the long run, and anxious attachment in the therapist was found to have a negative effect on client-rated alliance over time.

Ligiero and Gelso (2002) examined the relationship between alliance as rated by therapists and supervisors and therapist attachment patterns as measured by the RQ in 50 therapists conducting long term psychoanalytic treatment. They found no relationship between therapist attachment patterns and alliance as rated by either the therapists or their supervisors. In combination with the mixed results of Dunkle and Friedlander (1996) and Sauer et al. (2003) on the same issue this suggests that the relationship between therapist attachment patterns and alliance might be less straightforward than the relationship between client attachment patterns and alliance. It could be that the asymmetrical roles of therapist and client imply that therapist personality characteristics are not brought into play to the same extent as client characteristics. The more equivocal findings might also be an effect of therapists being more similar than clients are in terms of attachment patterns, thus making effects of therapist attachment patterns more difficult to detect. However, further studies on the subject are needed before drawing any firm conclusions on the matter.

Rather than focusing on the therapeutic alliance, other studies have dealt with the more direct link between therapist attachment patterns and therapist behaviour in therapy. Based on theory, one would expect secure therapists to show more sensitivity to clients, whereas therapists with insecure attachment patterns might be characterized by varying degrees of countertransferential behaviour originating in their own attachment-related issues. The previously mentioned study by Rubino et al. (2000) evaluated the attachment patterns of the 77 therapists-in-training by the RSQ and examined the relation between therapist attachment patterns and degree of empathy and depth of intervention in response to the simulated rupture episodes. RSQ scores were analyzed in terms avoidance and anxiety (Brennan et al., 1998), and degree of anxiety in the therapist was found to be negatively related to degree of empathy in responding to the rupture episodes. This is understandable, as the more anxious therapists might feel threatened by the rupture and consequently fail to empathize with the client. However, one would also expect therapist avoidance to affect response modes, but no such effects were found.

The above-mentioned study by Ligiero and Gelso (2002) also inquired into the relation between therapist attachment patterns and therapist countertransference behaviours as rated by supervisors. The assessed countertransference behaviours encompassed a number of antitherapeutic therapist behaviours such as rejecting the client, or talking too much in the session. Scores on the three insecure attachment patterns did not predict differences in countertransference behaviour, but there was a significant inverse relation between level of secure attachment in the therapist and negative countertransference behaviours. Even though one would expect scores on the insecure attachment patterns to predict a greater level of countertransference behaviour, the relationship that Ligiero and Gelso did find evidence of is consistent with attachment theory, in that one would expect secure therapists to be more able to refrain from engaging in negative countertransference behaviour.

Finally, the link between therapist attachment patterns and therapist behaviour in treatment was examined by Dozier, Cue, and Barnett (1994), who used the AAI scored with Kobak's Q-sort to evaluate the attachment patterns of 18 case managers working with 27 psychiatric patients. Case managers' descriptions of their interventions with patients were rated by observers for depth of intervention and attention to dependency needs in the patient. The score for depth of intervention distinguished between psychological interventions on one end of the scale and more practical help on the other end. More deactivating case managers were found to intervene in less depth and perceive less dependency needs in the patient, and more hyperactivating case managers were found to intervene in greater depth and perceive greater dependency needs in the patient.

Thus, case managers who scored high on deactivation or hyperactivation of attachment seemed prone to act accordingly. This finding seems to support the idea of countertransferential reactions in therapists who are insecure or lean towards one of the insecure attachment patterns.

Interaction effects between client and therapist attachment patterns

As both client and therapist attachment patterns seem to be predictive of some aspects of the therapeutic process, it is natural to inquire how they interact to produce combined effects on the therapeutic relationship. Such effects are obviously more difficult to research and document than simple effects of either client or therapist attachment patterns. Only four studies have assessed the attachment patterns of both clients and therapists. The already mentioned study by Sauer et al. (2003) did not find any interaction effects, whereas the study by Rubino et al. (2000) did find a trend for the more anxious therapists to be particularly unempathic with the fearful and secure clients compared to the dismissing and preoccupied clients. This is understandable for the fearful clients, as therapists that are more anxious may have trouble handling clients with similar attachment-related anxieties, but why an anxious therapist should be more prone to fail in empathy with the secure clients is not clear. However, the fact that the study was not conducted in the context of actual therapy with real clients makes the ecological validity limited.

The most detailed documentation of interaction effects between client and therapist attachment patterns is found in two studies by Dozier and colleagues. In the previously mentioned study by Dozier et al. (1994) the secure-anxious dimension in the attachment patterns of 18 psychiatric case managers predicted their response to patients with different attachment patterns. The insecure case managers intervened in greater depth and perceived greater dependency needs in patients who were more hyperactivating compared to patients who were more deactivating. The opposite was true of the secure case managers. This finding has some interesting implications with respect to complementarity in treatment. Hyperactivating patients would generally be expected to present themselves as more dependent and to "pull for" deeper interventions than deactivating patients do. The more insecure the case managers were, the more prone they were to react in accordance with this immediate presentation of the client. However, the secure case managers seemed instead to react in a way, which corrected for the insecure attachment patterns of the client (Bernier & Dozier, 2002).

Tyrrell, Dozier, Teague, and Fallot (1999) extended these results by an investigation of the effect of the attachment patterns of 54 clients and their 21 case managers upon therapeutic alliance. Attachment patterns were measured by the AAI scored with Kobak's Q-sort, and whereas most clients were rated as insecure, virtually all case managers were rated as secure, so the results were primarily analyzed in terms of differences on the deactivation—hyperactivation dimension. It was found that less deactivating case managers formed stronger alliances with more deactivating clients than with less deactivating clients. Furthermore, there was a nonsignificant trend for more deactivating case managers to form weaker alliances with more deactivating clients than with less deactivating clients.

Whereas the Dozier et al. (1994) study points to a tendency for secure therapists to spontaneously adopt a noncomplementary or corrective intervention style, the Tyrrell et al. (1999) study indicates that therapeutic alliances seem to be enhanced when client and therapist attachment patterns are incidentally combined in a noncomplementary way. Thus, therapists who are naturally inclined to act in a more deactivating or hyperactivating manner might have a special advantage in working with clients with the opposite tendency, presumably because of their ability to challenge or balance the attachment pattern of the client. However, so far the studies by Dozier and colleagues are the only studies to find this kind of match effect, while the other studies looking for such effects have failed to find them. Additional naturalistic studies on other populations than the psychiatric case management samples used in the studies by Dozier and colleagues would be useful to investigate the issue further.

Measurement of attachment patterns in the context of the therapeutic relationship

A somewhat different approach than assessing the general attachment patterns of clients and therapists and looking for correlates in the therapy setting is to try to measure the attachment that the client develops to the therapist directly. Mallinckrodt, Gantt, and Coble (1995) have developed the Client Attachment to Therapist Scale (CATS), which is a self-report measure distinguishing between secure, avoidant-fearful, and preoccupied-merger attachment to the therapist. Further studies using CATS suggest that it taps into important aspects of the therapeutic relationship (Mallinckrodt, Porter, & Kivlighan, 2005; Woodhouse, Schlosser, Crook, Ligiero, & Gelso, 2003). Similarly, but using a different method, Diamond and colleagues have devised the Patient—Therapist Adult Attachment Interview (PT-AAI), which is based on the format of the AAI and uses the same classificatory scheme (Diamond et al., 1999; Diamond, Clarkin, et al., 2003; Diamond, Stovall-McClough, Clarkin, & Levy, 2003). Whereas the CATS is only applicable to clients, the PT-AAI can be used with both clients and therapists.

Despite measuring attachment patterns within a similar conceptual framework, both the CATS and the PT-AAI correlate only partially with general measures of attachment patterns, stressing the importance of distinguishing between general attachment patterns and the pattern of attachment developed in specific new relationships. This also makes it difficult to integrate findings using these instruments with other findings described so far, since attachment patterns as measured by these instruments are not independent variables that can be related to psychotherapy process, but are themselves products of psychotherapy process. However, the measures clearly merit attention and further clarification of their relation to measures of general attachment patterns will shed more light on how the general attachment-related feelings and attitudes of clients and therapists are crystallized into unique new attachment relationships.

Studies of the impact of attachment patterns on the outcome of psychotherapeutic interventions are still quite modest in number and have primarily been concerned with predicting outcome on the basis of client attachment patterns. Client attachment patterns can be relevant to psychotherapy outcome in at least two ways. Firstly, clients with certain attachment patterns might fare better in psychotherapy or in particular forms of psychotherapy than other clients. Secondly, because of the link between insecure attachment patterns and psychological problems, it might be interesting to look at psychotherapy outcome in terms of change in attachment patterns. That is, adult attachment classification might be used as an outcome measure in itself.

The impact of attachment patterns on psychotherapy outcome

Clients with secure attachment patterns can be expected to engage more readily in a positive working relationship with a therapist, and one might consequently expect secure clients to have better therapeutic outcomes. However, since most clients are likely to be classified as insecure, differences in outcome between the insecure attachment patterns would obviously also be of interest. Fonagy et al. (1996) conducted a major study of inpatient psychoanalytically oriented treatment in a sample of 82 nonpsychotic patients and found that dismissing patients as assessed by the AAI showed greater improvement compared to secure and preoccupied patients. The explanation that the authors suggest for this somewhat surprising finding is that dismissing individuals, who have previously refrained from thinking of their problems in terms of the influence of past relationships, are more accessible to psychoanalytically oriented treatment than preoccupied individuals, who already have very well-formed and self-serving ideas about the influence of the past. This, however, does not explain why dismissing patients did better than secure patients, who were, however, small in number, and might not have been typical of secure patients in general.

Contrary to the results of Fonagy et al. (1996), a study by Horowitz, Rosenberg, and Bartholomew (1993) on 36 patients in brief dynamic psychotherapy indicated that dismissing patients had the poorest outcome. However, in this study the attachment patterns of patients were not assessed directly, but were inferred based on scores on the Inventory of Interpersonal Problems (IIP). Thus, the study showed that interpersonal problems falling in the "dominating", "vindictive", and "cold" octants of the IIP improved the least in the course of therapy. These interpersonal problems were in a separate study shown to be most typical of individuals with a dismissing attachment style as assessed by rating scales developed by Bartholomew and Horowitz (1991).

Meyer, Pilkonis, Proietti, Heape, and Egan (2001) investigated the relation between attachment patterns and outcome in 149 psychiatric patients in mixed psychiatric treatment. The attachment patterns of patients were assessed by rating intake interviews on a set of Attachment Prototypes developed by Pilkonis (1988). Pilkonis' Attachment Prototypes are similar to other measures of adult attachment in distinguishing between secure and insecure attachment patterns, but differ in terms of the number and characteristics of insecure attachment patterns. So far, they have not been compared with the more established measures of adult attachment patterns described in this article. The clearest result of the Meyer et al. study with respect to attachment patterns and outcome was a significant relation between the secure attachment prototype and a positive treatment outcome, which corresponds to the theoretical prediction.

These three studies of differences in outcome depending on client attachment patterns are difficult to compare, since they employed different measures of adult attachment, used different measures of therapeutic outcome, and were conducted on different client populations in different therapeutic modalities. Those differences aside, however, another reason for the lack of consistent findings in the above-mentioned studies might be that clients with different attachment patterns benefit differently from different forms of psychotherapy. Thus, one would not necessarily expect a particular attachment group to improve the most regardless of the type of treatment studied.

The potential existence of such differential match effects is suggested by the previously mentioned study by Tyrrell et al. (1999), which focused on the match between client and therapist attachment patterns rather than between client attachment pattern and type of treatment. In this study, the attachment patterns of 54 psychiatric outpatients and their 21 case managers were assessed with the AAI scored with Kobak's Q-sort. The results in terms of outcome measured by general assessment of functioning and patient ratings of life satisfaction indicated that more deactivating patients functioned better and were more satisfied when working with more hyperactivating (but secure) case managers, whereas more hyperactivating patients functioned better and were more satisfied when working with more deactivating (but secure) case managers. Thus it seems that clients benefit from working with a therapist with a noncomplementary interpersonal style, who can be expected to challenge the client's usual interpersonal style to a greater degree (Bernier & Dozier, 2002).

To the extent that one might describe different forms of psychotherapy as leaning towards a more deactivating or hyperactivating style, a similar pattern of matching might apply here. For instance, one might argue that cognitive behavioural psychotherapy is a more "deactivating" kind of treatment—not paying much attention to the relationship to parents or to the client—therapist relationship, but focusing instead on symptoms. Psychodynamic or interpersonal therapy might instead be considered a more "hyperactivating" kind of treatment—emphasising and elaborating on the importance of feelings and relationships. Some preliminary empirical support for match effects at the level of treatment types can be found in an unpublished study of cognitive and interpersonal group therapy for binge eating disorder, which showed that preoccupied clients as assessed by the AAI improved more with cognitive behavioural psychotherapy than with interpersonal psychotherapy. However, dismissing patients did equally well in both treatments (Borman Spurrell, 1996).

More research is clearly needed to draw any firm conclusions. If anything, the existing studies suggest that researchers will do well to formulate more fine-grained hypotheses, rather than trying to establish which adult attachment group that does best on average. This could also make studies more directly relevant to clinical decision making. For instance, it might be useful to distinguish between psychotherapy targeting interpersonal problems directly and psychotherapy targeting other more specific difficulties such as for instance eating disorders. Attachment patterns of clients probably play a different role in treatment depending on whether they are part of the intervention focus or rather a mediating or moderating variable affecting the treatment of other specific difficulties.

Changes in attachment patterns as an outcome of psychotherapy

In view of the relation between insecure attachment patterns and interpersonal and psychological difficulties, it seems natural to inquire whether psychotherapy is capable of changing client attachment patterns. In the previously mentioned study by Fonagy et al. (1996), psychiatric patients were administered the AAI before and after one year of psychoanalytically oriented inpatient treatment. In a different publication, the authors report on change in AAI status of 35 of the total 82 patients, stating that whereas all of those 35 patients were insecure upon admission, 40% of the patients were classified as autonomous on discharge (Fonagy et al., 1995). In spite of this rather impressive result, a more detailed account of the changes in AAI status observed in this study has not been published to date.

Currently the AAI is similarly being employed as an outcome measure in a large study of Transference Focused Psychotherapy for borderline personality disorder (Diamond et al., 1999; Diamond, Clarkin, et al., 2003; Diamond, Stovall-McClough, et al., 2003). To date data on attachment patterns at the beginning and after one year of therapy have been published for 10 patients, and the results are rather complex. Three patients changed from insecure to secure patterns, four patients did not show much change, and three patients changed from classifiable insecure states to the category cannot classify, which might indicate some kind of ongoing reorganization, but is normally considered more pathological. Four patients initially classified as unresolved were not classified as such following treatment. A more uniform picture is suggested by data on reflective functioning, where the majority of the patients showed improvements. Generally, it seems that the results reported by Diamond, Clarkin, et al. (2003) and Diamond, Stovall-McClough, et al. (2003) are somewhat less optimistic than the 40% change to security reported by Fonagy et al. (1995), but analysis of the total material must be awaited before anything can be concluded in this regard.

Travis, Bliwise, Binder, and Horne Moyer (2001) studied change in attachment patterns as assessed by rating scales developed by Bartholomew and Horowitz (1991) during the course of time-limited dynamic psychotherapy. They report that in a group of 29 clients who were all rated as insecure before treatment, there was a significant increase in the number of clients with secure attachment and a significant decrease in the number of clients with fearful attachment. Sixty-six percent of the clients changed attachment style over the course of therapy, but the majority of those changed from one insecure attachment style to another. The clients who changed to a secure attachment style (24%) were found to be the ones exhibiting the lowest level of symptoms before treatment, but they had the same rate of symptom reduction during treatment as the other clients.

There are two important caveats in interpreting the results of studies showing changes in attachment patterns after psychotherapy. First, one has to consider the test—retest reliability of the measures used to assess attachment patterns, and the expected stability of adult attachment patterns in untreated groups. A portion of the observed change is always attributable to measurement error, and furthermore adult attachment patterns do change over time, and are particularly unstable in high-risk and clinical populations (Waters et al., 2000). Thus, changes in attachment patterns during the course of psychotherapy cannot be confidently attributed to treatment effects without comparison to a control group. Still, the existing studies mostly showed a shift in the direction of greater security rather than the more random change that would be expected if only measurement error and general instability of attachment patterns were in play. In addition, secure attachment is actually quite rare in clinical groups, making the posttreatment rates of secure attachment noteworthy. Psychotherapeutic treatment thus seems to be able to shift adult attachment patterns in the direction of greater security.

This, however, leads to the second caveat, namely that the meaning of a change in attachment pattern as assessed by interview or self-report is not completely straightforward. There might conceivably be change in a client's attachment-related concepts of self or other or in a client's ability to describe his childhood in a nuanced and coherent manner without much accompanying change in the client's real life relationships (Eagle, 1997). To validate the genuineness of changes in attachment patterns, studies using attachment patterns as an outcome variable should add other instruments targeting the aspects of client functioning hypothesized to change as a result of a change in attachment patterns.

Even though research on adult attachment patterns in individual psychotherapy is still in its infancy, the growing numbers of empirical studies clearly support the relevance of adult attachment to psychotherapy process and outcome. The existing studies have linked the attachment patterns of clients and therapists to differences in in-treatment behaviour, to differences in the quality and development of the therapeutic alliance, and to differences in therapeutic outcome. However, the field is marked by a plethora of different measuring instruments and research tactics, which makes it rather difficult to compare studies and extract the essence of the empirical findings so far. Thus, progress in research on adult attachment patterns and psychotherapy is very much dependent on advances in the general field of adult attachment. A greater consensus on the specific meaning of the concept of adult attachment patterns, a greater consensus on relevant measuring instruments and their interrelations, and an increased knowledge of what scores on these instruments predict outside the realm of therapy would be helpful to psychotherapy researchers.

As with other fields within psychotherapy research, clinicians might relevantly ask about the use value of the empirical findings on adult attachment patterns and individual psychotherapy. Here the issue of assessment becomes essential, as clinicians will need quick and reliable assessment of client attachment patterns to be realistically able to consider them in their choice of intervention strategies. Even without formally assessing client attachment patterns, however, a knowledge of what characterizes the different attachment patterns can probably serve as an inspiration to clinicians. Parts of the attachment literature put a lot of emphasis on the merits of attachment security, and the concept of security can obviously be an important guide in clinical work. However, the knowledge that people with secure attachment patterns are easier to treat and benefit more from treatment is not of much use to clinicians when the majority of clients actually have insecure attachment patterns. Thus from a practical viewpoint, future research on adult attachment patterns in individual psychotherapy could profitably inquire more into the challenges of treating people with different insecure attachment patterns, rather than comparing how secure and insecure clients behave and benefit from treatment.