Skip to content
Psych Dictionary

Adult Attachment Interview

(AAI)

  • Method
  • Moderate evidence
  • Published October 19, 2026
  • 14 sources
  • 5 min read

In short

The Adult Attachment Interview is a semi-structured interview, developed at Berkeley in the 1980s by Carol George, Nancy Kaplan and Mary Main, that classifies an adult’s state of mind with respect to attachment from the coherence of the account of childhood, not from what happened.

Key facts

Developed by
Carol George, Nancy Kaplan and Mary Main (interview protocol); Mary Main, Ruth Goldwyn and Erik Hesse (coding system)
Origin
1982–1985, University of California, Berkeley
Also known as
AAI; Berkeley Adult Attachment Interview

State of the evidence

Moderate evidence

Classifications were stable over two to three months in two test–retest studies and were independent of memory for non-attachment events and of intelligence[1][2]. A 1995 meta-analysis of 18 samples found that parents’ classifications predicted their infants’ Strange Situation security with a large effect[3], but a synthesis of 95 samples found smaller effects (r = .31), smaller still in recent and unpublished studies[4]. Classifications barely agree with self-report attachment measures[5], the secure–dismissing distinction appears dimensional rather than categorical[6], and the coding manual remains unpublished[7].

Evidence labels summarise the research literature; they are not recommendations. How labels are assigned

The Adult Attachment Interview (AAI) is a semi-structured interview in which an adult describes childhood relationships with attachment figures and is then classified by how the account is given. It was designed to assess adults’ current mental representation of their childhood attachment experiences[2], and it prompts speakers to retrieve and evaluate attachment-related memories[1]. Transcripts are coded for the speaker’s capacity to discuss those experiences coherently, and a person with an unfavorable history can be classified secure if the history is recounted coherently[8]. Carol George, Nancy Kaplan and Mary Main developed the interview[7], and van IJzendoorn dated its introduction to 1985[3].

The AAI is not a questionnaire, and it does not measure the adult romantic attachment styles assessed by self-report scales; the relation between the two traditions is set out in that entry. Nor is it freely available. The scoring and classification manual is unpublished, early versions of the protocol are cited from manuscripts, and access to the coding system has come mainly through in-person training[7].

Origins in the Berkeley study

The interview grew out of a six-year follow-up of infants whom Main and colleagues had seen in Mary Ainsworth’s Strange Situation. While the children completed tasks, their mothers and fathers were interviewed separately about their own attachment experiences[7]. John Bowlby had drawn on Tulving’s distinction between episodic memory, of particular events, and semantic memory, of generalized propositions, and the interview was built to elicit both[7].

Coding was developed inductively. Ruth Goldwyn and Main compared parents’ transcripts with the Strange Situation classifications their infants had received five years earlier, and found distinctive ways of speaking associated with each infant pattern. Parents in avoidant dyads often described flatly ideal childhoods that they could not support with specific memories[7]. During a visit to Berkeley in 1982, Mary Ainsworth and Main concluded that these idealized accounts relied on semantic memory at the expense of episodic memory[7]. Main, Kaplan and Cassidy set out the approach in 1985 as a move to the level of representation[8][7].

Structure of the interview

As Duschinsky describes the early protocol, the interview opens by asking the speaker to describe the relationship with the parents in childhood. It then asks for five adjectives to describe each relationship, a prompt for semantic memory, followed by specific memories that illustrate each adjective, a prompt for episodic memory[7]. Later questions ask what the speaker did when upset as a child, whether parents held them for comfort, and whether they felt rejected, with an account of why the parents acted as they did. Further questions cover major separations, occasions when parents were threatening, experiences of loss, and how early experiences have affected the speaker as an adult[7]. The questions follow one another with little conversational reciprocity from the interviewer[7]. Interviews have been given in pregnancy as well as after a child’s birth[9].

Coding states of mind

The coding system examines the transcript as a whole rather than the answers to particular questions. It attends neither to the speaker’s stated opinions nor to the speaker’s actual history, which an interview cannot establish, but to how the speaker attends to and communicates about attachment-related experience[7]. Main describes the analysis as an assessment of the speaker’s capacity to adhere to Grice’s maxims of rational, cooperative discourse[8]. In Duschinsky’s summary, these are quality (be truthful, with evidence for what is said), quantity (be succinct yet complete), relation (be relevant) and manner (be clear and orderly)[7].

Secure-autonomous

The speaker describes positive and negative aspects of early relationships with balance, draws on both semantic and episodic memory, and stays coherent and collaborative regardless of the difficulty of the subject[7].

Dismissing

The speaker minimizes negative experiences or their importance, offers generalizations, often idealizing, that specific memories do not support, and gives answers that are too brief[7].

Preoccupied

The speaker remains angrily or passively absorbed in past relationships, taking long conversational turns that lose track of the questions[7].

Unresolved/disorganized

Marked lapses in reasoning or discourse while the speaker discusses loss or abuse; this classification predicts disorganization in the speaker’s infant[10].

Cannot classify

A global breakdown in the organization of a single strategy for meeting the discourse tasks of the interview[11].

The first three are the organized categories[8]. The unresolved code is given in addition to one of them[12], on a nine-point scale on which scores above 5 lead to the classification[7]. By convention, transcripts classified as cannot classify are combined with unresolved transcripts in statistical analyses[7].

Reliability and discriminant validity

Two early studies tested whether classifications were stable and whether they reflected something other than memory or intelligence. In one, 83 mothers were interviewed twice, two months apart, by different interviewers. Their classifications agreed in 78% of cases on the three main categories (κ = .63), with the unresolved category less stable, and they were independent of memory for non-attachment events, of verbal and performance intelligence and of social desirability[1]. In the second, 59 Israeli students interviewed twice at three-month intervals showed high interjudge and test–retest reliability, irrespective of interviewer, and no association with non-attachment memory or intelligence[2]. Over twelve months, mothers’ classifications were stable in 90% of cases with three categories and 77% with four[9].

The categories may not mark natural types. A taxometric analysis of 504 interviews found variation between secure and dismissing states of mind more consistent with a dimension than with distinct groups; the analysis of preoccupation was indeterminate[6].

Prediction across generations

The interview’s main claim to validity is that a parent’s classification predicts the child’s attachment to that parent. In a 1995 meta-analysis of 18 samples (N = 854), parents’ classifications predicted infants’ security in the Strange Situation with a combined effect of d = 1.06. In the samples where correspondence could be computed, secure and insecure classifications matched in 75% of cases (κ = .49), and classifications also predicted parental responsiveness (d = .72)[3]. Interviews given in pregnancy predicted infants’ later Strange Situation classifications in 81% of cases with three categories, and grandmothers’ classifications predicted their grandchildren’s in 75%[9].

Later results were smaller. A synthesis of 95 samples (N = 4,819) found transmission of r = .31 for secure-autonomous states of mind and r = .21 for unresolved ones, against .47 and .31 two decades earlier. Effects were smaller in unpublished and recent studies and varied with sample risk, biological relatedness and the child’s age[4]. Parental sensitivity explains part of the link but not all of it, the shortfall known as the transmission gap[3][4]. The gap is wider for disorganization: anomalous parental behavior accounted for only a small part of the association between unresolved states of mind and disorganized attachment in infants[13]. The developmental account of these findings is given under attachment theory.

Earned security

Some speakers who describe difficult childhoods are nonetheless classified secure-autonomous, a group Main came to call earned-secure by 1988[7]. A 23-year longitudinal study tested whether such speakers had in fact had insecure early relationships. Speakers defined retrospectively as earned-secure were no more likely than continuous-secure speakers to have been insecurely attached as infants. In childhood and adolescence they had been observed to receive among the most supportive maternal care in a high-risk sample[14]. The taxometric analysis likewise found no qualitative distinction between earned-secure and continuous-secure speakers[6].

What the interview measures

Self-report attachment questionnaires and the AAI agree only trivially, with a mean correlation of .09 across ten studies[5]. The question of what the interview does measure has its own history. Main first linked the AAI to Bowlby’s internal working models, but from 1987 she stopped equating states of mind with internal working models and later called that characterization misleading[7]. In Main and Hesse’s later conception, as Duschinsky reports it, the interview measures how a person deploys attention and communication about attachment-related experience, not the attachment system itself[7].

The technical sense of coherence also differs from everyday usage. In one study, coherence as coded in the AAI correlated only .37 with a conventional measure of narrative coherence[7]. Because the manual and the principles behind it have not been published, the coding system has been less exposed to technical criticism and psychometric appraisal by researchers outside Main’s network[7].

Sources14

Numbered in order of first citation.

  1. 1.Bakermans-Kranenburg, M. J., & Van IJzendoorn, M. H. (1993). A psychometric study of the Adult Attachment Interview: Reliability and discriminant validity. Developmental Psychology, 29(5), 870–879. https://doi.org/10.1037/0012-1649.29.5.870
  2. 2.Sagi, A., van IJzendoorn, M. H., Scharf, M., Koren-Karie, N., Joels, T., & Mayseless, O. (1994). Stability and discriminant validity of the Adult Attachment Interview: A psychometric study in young Israeli adults. Developmental Psychology, 30(5), 771–777. https://doi.org/10.1037/0012-1649.30.5.771
  3. 3.van IJzendoorn, M. H. (1995). Adult attachment representations, parental responsiveness, and infant attachment: A meta-analysis on the predictive validity of the Adult Attachment Interview. Psychological Bulletin, 117(3), 387–403. https://doi.org/10.1037/0033-2909.117.3.387
  4. 4.Verhage, M. L., Schuengel, C., Madigan, S., Fearon, R. M. P., Oosterman, M., Cassibba, R., Bakermans-Kranenburg, M. J., & van IJzendoorn, M. H. (2016). Narrowing the transmission gap: A synthesis of three decades of research on intergenerational transmission of attachment. Psychological Bulletin, 142(4), 337–366. https://doi.org/10.1037/bul0000038
  5. 5.Roisman, G. I., Holland, A., Fortuna, K., Fraley, R. C., Clausell, E., & Clarke, A. (2007). The Adult Attachment Interview and self-reports of attachment style: An empirical rapprochement. Journal of Personality and Social Psychology, 92(4), 678–697. https://doi.org/10.1037/0022-3514.92.4.678
  6. 6.Roisman, G. I., Fraley, R. C., & Belsky, J. (2007). A taxometric study of the Adult Attachment Interview. Developmental Psychology, 43(3), 675–686. https://doi.org/10.1037/0012-1649.43.3.675
  7. 7.Duschinsky, R. (2020). Cornerstones of attachment research. Oxford University Press.
  8. 8.Main, M. (2000). The organized categories of infant, child, and adult attachment: Flexible vs. inflexible attention under attachment-related stress. Journal of the American Psychoanalytic Association, 48(4), 1055–1096. https://doi.org/10.1177/00030651000480041801
  9. 9.Benoit, D., & Parker, K. C. H. (1994). Stability and transmission of attachment across three generations. Child Development, 65(5), 1444–1456. https://doi.org/10.2307/1131510
  10. 10.Hesse, E., & Main, M. (2000). Disorganized infant, child, and adult attachment: Collapse in behavioral and attentional strategies. Journal of the American Psychoanalytic Association, 48(4), 1097–1127. https://doi.org/10.1177/00030651000480041101
  11. 11.Hesse, E. (1996). Discourse, memory, and the Adult Attachment Interview: A note with emphasis on the emerging cannot classify category. Infant Mental Health Journal, 17(1), 4–11. https://doi.org/10.1002/(SICI)1097-0355(199621)17:1<4::AID-IMHJ1>3.0.CO;2-S
  12. 12.Bakermans-Kranenburg, M. J., & van IJzendoorn, M. H. (2009). The first 10,000 Adult Attachment Interviews: distributions of adult attachment representations in clinical and non-clinical groups. Attachment & Human Development, 11(3), 223–263. https://doi.org/10.1080/14616730902814762
  13. 13.Madigan, S., Bakermans-Kranenburg, M. J., Van Ijzendoorn, M. H., Moran, G., Pederson, D. R., & Benoit, D. (2006). Unresolved states of mind, anomalous parental behavior, and disorganized attachment: A review and meta-analysis of a transmission gap. Attachment & Human Development, 8(2), 89–111. https://doi.org/10.1080/14616730600774458
  14. 14.Roisman, G. L., Padrón, E., Sroufe, L. A., & Egeland, B. (2002). Earned–secure attachment status in retrospect and prospect. Child Development, 73(4), 1204–1219. https://doi.org/10.1111/1467-8624.00467

Further reading

  • Duschinsky, R. (2020). Cornerstones of attachment research. Oxford University Press.Open access (CC BY-NC-ND 4.0).
  • van IJzendoorn, M. H. (1995). Adult attachment representations, parental responsiveness, and infant attachment: A meta-analysis on the predictive validity of the Adult Attachment Interview. Psychological Bulletin, 117(3), 387–403. https://doi.org/10.1037/0033-2909.117.3.387
  • Verhage, M. L., Schuengel, C., Madigan, S., Fearon, R. M. P., Oosterman, M., Cassibba, R., Bakermans-Kranenburg, M. J., & van IJzendoorn, M. H. (2016). Narrowing the transmission gap: A synthesis of three decades of research on intergenerational transmission of attachment. Psychological Bulletin, 142(4), 337–366. https://doi.org/10.1037/bul0000038

Cite this entry

  • APA 7

    Psych Dictionary. (2026, October 19). Adult Attachment Interview (AAI). In Psych Dictionary. https://www.psychdictionary.com/methods/adult-attachment-interview

  • MLA 9

    “Adult Attachment Interview (AAI).” Psych Dictionary, 19 Oct. 2026, www.psychdictionary.com/methods/adult-attachment-interview.

Written by Editorial team. First published October 19, 2026. Editorial standards

Educational reference material. Nothing on this site is a diagnosis, a treatment recommendation or a substitute for care from a qualified clinician.