Updated on
August 27, 2026
A clear comparison of attachment theories, research methods and classifications, from Bowlby and Ainsworth to adult models, with evidence limits for teachers and scholars.
What are the main theories of attachment?
Attachment research is best understood as one broad Bowlby-Ainsworth tradition with later models and measures, not as four equivalent theories. Bowlby proposed an evolved attachment behavioural system. Ainsworth developed observations and the Strange Situation, which produced infant classifications. Main and colleagues studied attachment representations through the Adult Attachment Interview. Adult social psychologists later developed romantic-relationship categories and anxiety-avoidance dimensions. These methods measure different relationships, ages and constructs, so their labels cannot be swapped or inferred from ordinary classroom behaviour.
Attachment research is one broad Bowlby-Ainsworth tradition, not four rival theories. Bowlby set out a system for how a young child seeks safety. Ainsworth devised close field work and the Strange Situation infant test. Main and her peers later studied how adults speak about past bonds.
A separate adult field used surveys about close or romantic bonds. These tests cover distinct ages, ties and ideas. Their labels cannot be swapped. Nor can a teacher infer them from day-to-day acts in class.
A sound summary asks five things about each claim: the theory, test, person or bond, result and limit. This takes more care than a four-style quiz. It is also much more useful for teachers, SENCOs and scholars. Readers new to this field may first want our guide to the main theories of learning.
The main account is Bowlby's ethological theory of attachment. Ainsworth's field work helped test it. Earlier learning and psychoanalytic accounts gave other reasons for how bonds form. Each account must stay tied to its own age, bond and test.
Later work added infant groups, adult interviews and self-report models. These strands are linked, but they do not test the same thing.
Early learning accounts asked if food and rewards could explain an infant's bond. Bowlby gave a different view. He said natural selection shaped a goal-led system. It prompts a child to seek a known carer when scared, then to explore when safe (Bowlby, 1969).
Ainsworth helped turn parts of the theory into a test. Her home studies and Strange Situation examined how an infant used a known carer while at play, apart and back together. The first work gave three organised groups: avoidant, secure and resistant or ambivalent (Ainsworth et al., 1978).
Main and Solomon later set out codes for disorganised or disoriented acts in the Strange Situation. Main, Kaplan and Cassidy then studied how people speak about bonds (Main et al., 1985). This work grew from the infant field, but its test and codes mean something else.
Adult social psychologists took another route. Hazan and Shaver (1987) used attachment ideas to study love. Bartholomew and Horowitz (1991) tested a four-group adult model. Later surveys often used two scales: anxiety and avoidance (Fraley et al., 2000).
For a fuller account of the behavioural system, secure base, safe haven, monotropy and internal working models, use the dedicated guide to Bowlby's attachment theory. This page owns the comparison across models and methods.
A theory explains a process. A method is the way a study tests it. A code or score is one result from that method. A finding is what one study or review found.
Keep these terms apart. This stops a lab group from turning into a trait label. It also stops a link between two things from being called a cause.
On a small screen, swipe the table horizontally to compare all columns.
| Contribution | What it is | Typical output | What it does not establish |
|---|---|---|---|
| Bowlby's attachment system | A theory about proximity, protection, exploration and internal working models | Testable propositions about behaviour and relationships | An individual classification from ordinary observation |
| Strange Situation | A structured infant-caregiver observation procedure | A, B or C classification, with later D coding | A diagnosis, school-age trait or adult romantic style |
| Adult Attachment Interview | A structured interview coded for discourse about attachment experiences | State-of-mind classifications and rating scales | A direct measure of romantic behaviour or the quality of childhood care |
| Romantic self-report | A questionnaire about thoughts and feelings in close adult relationships | Categories or anxiety and avoidance scores | An infant Strange Situation classification |
| Outcome study | A study of association, prediction or change | An effect estimate with uncertainty | Determinism, diagnosis or one cause for an individual outcome |
This rule changes how a teacher reads a claim. “The learner avoids help” notes what took place. It is not an attachment code. The same act may stem from the task, speech, past feedback, fear, sensory load, culture or the bond in that lesson.
The same rule matters for scholars. An Adult Attachment Interview result cannot be merged with a survey score for anxiety. The tests use different prompts, codes and schools of thought. Fraley and Roisman (2014) studied groups and scales within AAI ratings, not all attachment tests.
Early accounts linked the bond to food, drive loss, learned cues or ideas from psychoanalysis. Bowlby drew on that last field but did not think food was enough. Work on animal life and control systems helped him frame a goal-led system for safety and closeness.
One learning account said that food cuts a basic drive such as hunger. The carer, paired with food, then gains reward value. Other accounts asked how a carer's cues become tied to ease, less distress or approach. Their focus was what the child learned from cues and results.
Rajecki et al. (1978) compared ethological, operant and conditioning accounts of social bonds. The review shows what the phrase once meant. It named rival accounts of how a bond forms, not a list of secure, anxious, avoidant and disorganised labels.
Bowlby saw the search for closeness as part of an evolved system, not a side effect of food. Young animals and infants seek touch when no food is on offer. This challenged a food-only account.
It did not show that learning has no role. What a child has lived through can still shape trust and acts.
Psychoanalytic ideas shaped Bowlby's work on loss, time apart and mental models. Yet his later account put real bonds and acts at its heart. He drew from more than one field rather than just using an old school of thought.
In teacher education, this distinction prevents an exam-style list from becoming a practice rule. A lecturer can ask learners to compare the mechanism proposed by learning theory with Bowlby's ethological mechanism. A school should not use the resulting labels as a checklist for sorting children.
Bowlby said that acts linked to attachment form a goal-led system. Threat, illness, time apart or doubt can switch it on. The child then seeks a known person who offers care.
When the child feels safe, play and search become more likely. Past bonds can also shape what the child expects from self and others.
One visible act does not have one fixed meaning. Tears, a look, an approach or a move away must be read in context. Age, the bond and what set off the system all matter. Attachment refers to a bond, not a set amount of need or poise.
A secure base helps a child explore. A safe haven gives care when stress is high. These are roles in the theory, not praise for a child's traits.
A child may explore with ease in one place and need more help in the next. The acts we see can change with age and context.
Internal working models are the views and hopes built through past ties. Some guides turn them into scripts that fix each bond for life. Bowlby's account left room for new life events and new ties. Later work also finds both some link across time and scope for change.
A theory may be much wider than the test used for it. A Strange Situation code comes from a short set task with a known carer. It does not read a mental model at first hand. One event in class cannot reveal the whole bond system.
Bowlby's work sits within wider child-development theories. It does not rule out mood, speech, thought, SEND, family life, culture, teaching or stress. A good account keeps more than one cause in view and changes as new facts emerge.
Ainsworth paired close field notes with a set infant-carer test. The Strange Situation uses short spells of play, a stranger, time apart and reunion. Trained coders study the full pattern. They pay close heed to how the infant uses the carer when they meet again.
The first groups were A, B and C. Group B infants used the carer as a secure base and could settle after reunion. Group A showed more acts of avoidance at reunion. Group C showed resistant or mixed acts and found it hard to settle.
These are research groups from one set test. They are not broad child types.
Distress when apart does not set the code on its own. A child may be upset yet show a secure pattern at reunion. An infant may also look calm while other signs show stress. The full pattern and trained rules matter.
The test was made for infants in a set age band. Other tests exist for older groups, but the same act cannot be read in the same way. A school-age learner who looks away is not showing a Strange Situation A code. The bond, task and age all differ.
Ainsworth also did long field work outside the lab. Her work was not just a list of styles. It joined close notes on infant-carer contact with a set test. This let teams compare patterns in the same set-up (Ainsworth et al., 1978).
Use across cultures needs care. van IJzendoorn and Kroonenberg (1988) found gaps both across and within groups. A pooled rate is not a rule for one school, town or land. Local norms can shape how time apart, closeness and play look.
A teacher observing a learner during transition can still gather useful evidence. Note when the difficulty occurs, who is present, what demand changed and which support helped. That record serves classroom planning. It is not a substitute for an attachment assessment.
Main and Solomon set out codes for disorganised or disoriented acts in the Strange Situation. Main, Kaplan and Cassidy also used set adult talks to study mental models of bonds. These were new ideas and tests, not one more style to spot in class.
Coders use D when set acts in the Strange Situation show a clash or break in the bond system. It does not mean random, unruly or hard-to-guess acts in daily life. It is not an attachment disorder.
Duschinsky and Solomon (2017) say that three levels are often merged. These are the acts in a code, a break in the bond system and D as a group. They are not one thing. Daily use of the word “disorganised” adds to the mix-up.
Granqvist et al. (2017) correct common misuse. The infant code does not prove harm, illness, a fixed trait or an attachment disorder. Risk found in a group cannot tell us what took place for one child.
Main et al. (1985) moved to mental models. The Adult Attachment Interview asks adults about past bonds. Coders focus on how the person speaks, such as how clear and linked the account is.
Its groups concern state of mind about attachment. They do not score love style or judge if a parent was good or bad.
The AAI does not group an adult by a count of hard events. Two people with hard pasts may get different codes because the way they speak differs. A clear account does not mean the past was easy. The test needs expert skill and close coding.
Adult attachment work has two main test strands. One uses talks such as the AAI to study state of mind about past bonds. The other uses self-report forms about close or romantic ties. Both strands use some of the same words, but they do not test the same thing.
Hazan and Shaver (1987) used attachment ideas to study love. Their first survey used groups drawn from infant work. It led to a large field. It did not show that an infant code stays the same in an adult bond.
Bartholomew and Horowitz (1991) tested an adult model with four groups. It drew on good or poor views of self and others. The groups were secure, preoccupied, dismissing and fearful. Fearful adult attachment is not the same as the infant D code.
Later adult surveys often use two scales: anxiety and avoidance. Anxiety covers fear of loss, lack of care or being left. Avoidance covers unease with trust, closeness or help. The person gets scores, not a type for life.
Fraley et al. (2000) used item-response methods to improve adult self-reports. A scale keeps more of the range than one group label. Yet each score still depends on the words, bond, sample and aim of the form.
Adult self-reports are easy to give, so they are common online. A short quiz may feel true while it lacks sound scores, a named bond or fair norms. It cannot diagnose, fix fate or rebuild a true account of infancy.
For a learner writing an assignment, state the tradition before the label. Write “avoidant classification in the infant Strange Situation”, “dismissing state of mind on the AAI” or “higher avoidance on an adult self-report”. This small wording choice prevents a large conceptual error.
Attachment tests differ by age, bond, task, source and score. No one tool spans infancy, school, adult life and each close tie. A side-by-side view shows why one label cannot pass from a carer test to a teacher report or love survey.
On a small screen, swipe the table horizontally to compare all measures.
| Measure | Age and relationship | Evidence gathered | Typical output | Main boundary |
|---|---|---|---|---|
| Strange Situation | Infant with a familiar caregiver | Structured exploration, separation and reunion behaviour | A, B, C and later D coding | Not a school checklist or disorder diagnosis |
| Attachment Q-sort | Young child in a caregiving context | Extended observation sorted against a criterion | Security score and related descriptions | Needs trained observation and is relationship-specific |
| Child Attachment Interview | School-age child or young person | Structured interview and discourse | Representational classifications or ratings | Not inferred from teacher observation alone |
| Adult Attachment Interview | Adolescent or adult discussing attachment experiences | Transcribed narrative coded for discourse qualities | State-of-mind classification and scales | Not a romantic relationship questionnaire |
| Adult romantic self-report | Adult reporting on close or romantic relationships | Questionnaire responses | Categories or anxiety and avoidance scores | Not a retrospective infant classification |
NICE lists tools for young people who have, or may have, attachment needs. The tool must fit the child's age. NICE also says trained health and care staff should use it (NICE, 2015). This is far from a teacher matching acts to a wall chart.
The aim of a test matters. A study may seek gaps across groups. A health team may plan care. A school may need a clear log of stress, access, risk and what helped.
These jobs do not need the same tool. One should not borrow the force of the next.
Reports of warmth, strife and need have also been used to study teacher-child ties. Spilt and Koomen (2022) say these ties tend to have an attachment part, but are not full attachment bonds. A score for bond quality does not reveal a child's attachment style.
Use the narrowest sound words. “The learner sought help from this teacher in the task” is stronger than “the learner is anxiously attached”. The first can be checked and used in a plan. The second makes an unproved claim about a bond and past life.
Attachment tests show small to modest links with some later social and behavioural results. Some long-term studies still find links after they account for a set of prior facts. This means the bond can form one part of growth. It does not set a learner's fate or prove one cause of a school result.
Fearon et al. (2010) pooled links between insecure or disorganised attachment and acts aimed out at others. The result is about risk across groups.
Such acts do not reveal an attachment code. Nor do the data prove the main cause for one child.
Groh et al. (2014) found a modest link between security with a carer and skill with peers. A modest mean leaves wide overlap between groups. Children coded secure may find peer life hard. Those coded insecure may have strong peer ties.
Pinquart et al. (2013) found some link in attachment from infancy to early adult life. The size of the link changed with test, time and context. The result rejects both stark claims.
Attachment is not wholly random through time. Yet an early code is not a life term that can never change.
Shared causes and later life can shape a forecast. Care, stress, speech, health, SEND, mood, peers, school and social life may affect both the bond score and the later result. A link in data does not sort out all these paths.
On a small screen, swipe the table horizontally to compare evidence and interpretation.
| Evidence statement | Safe interpretation | Unsafe leap | School response |
|---|---|---|---|
| Group association with externalising behaviour | Attachment is one probabilistic risk marker in research | Disruption proves disorganised attachment | Assess needs, context and safety without typing the learner |
| Association with peer competence | Relationship history may be one contributor | Attachment determines friendships | Observe the specific interaction and teach relevant skills |
| Non-zero stability over time | Continuity and change both occur | Infancy fixes adult attachment | Avoid fatalistic language; attend to current relationships and support |
| Teacher-child relationship predicts outcomes | Relationship quality matters in a school context | The teacher replaces the caregiver or repairs attachment | Provide predictable, respectful teaching and use referral routes when needed |
Attachment work can sharpen questions about trust, felt safety and the search for help. It cannot replace checks on teaching, access to the work, emotion regulation, speech, health, SEND or risk. Use it as one lens, not the sole account.
The secure-base role appears in work from many settings. Yet acts and group rates vary across samples and cultures. Some links last through time and some change. Most results show a group trend, not one fixed cause.
Most outcome work tracks what occurs rather than running a true cause test. A sound summary keeps all four points in view.
The Strange Situation arose in one time and place. Time apart and reunion may mean different things in different care norms. Local views also shape closeness, self-rule and signs of fear. A test can be coded well yet still need care when we read the result.
van IJzendoorn and Kroonenberg (1988) found gaps both across and within cultures. A land label tells us little on its own. A school must not take a Western rate and assume it fits its own learners.
It is easy to overstate what stays the same. Tests change with age because acts, speech and bonds change. A later talk or survey is not the same test done again. A stable home may help explain a link through time, while new ties and life events can bring change.
Cause is hard to test because a team cannot assign an infant to a care past. Many studies adjust for a set of known facts, but some facts may still be missed. Trials can give firmer cause tests for one scheme. They do not prove each school plan sold under an attachment label.
A meta-analysis pools studies and can give a more exact mean. It does not erase gaps in tests, samples or bias. A mean can hide a wide range.
What gets run and shared also shapes the record. State the trend and its limits, not just “attachment affects learning”.
Attachment theory supports one modest school claim. Safe, kind and steady ties can help learners take part. The theory does not let staff diagnose a bond, rebuild a family past or match care to a guessed style. Staff should meet seen needs, guard access and safety, and seek expert help when needed.
Start with good practice for all. Greet learners but do not press them to share their past. Make plans and change clear. Teach ways to seek help.
Make feedback name the work and next step. Repair a mix-up. Keep rules calm and steady. These steps can help many learners and do not need an attachment code.
Then name the actual block. A learner who does not seek help may not know the routes on offer. They may need more time, a way to point, a small first step, some space or a clearer task. Let the block guide the help, not a guessed past.
Spilt and Koomen (2022) show that teacher-child ties can serve some bond roles, most of all in early years. They also say these ties are less deep, long and sole than care bonds. To call each teacher an attachment figure goes past both the data and the role.
Our guide to attachment theory in education gives the full school use. It covers sound practice for all, the limits of school accounts, referral and risk. This page keeps the models and tests clear.
Schools can apply inclusive education principles without waiting for a category. Clear teaching, communication access, safe participation, predictable support and reasonable adjustments are valuable regardless of why a learner currently finds a relationship or task difficult.
Most errors are leaps from one type of fact to the next. An act becomes a code. A code becomes a diagnosis. A group link is then used to explain one child's past or needs.
A link becomes a cause. An adult label is cast back on an infant. The fix is to name the test, bond, age and claim each time.
Better interpretation: Ainsworth and colleagues reported three organised Strange Situation classifications. Main and Solomon later developed disorganised or disoriented coding. Presenting all four as Ainsworth's theory erases the chronology and turns method outputs into a theory list.
Better interpretation: disorganised attachment is a specialist classification coded from specified behaviour in the infant Strange Situation. Granqvist et al. (2017) reject its use as a fixed trait, definitive maltreatment indicator or attachment disorder. Describe classroom events in their own terms.
Better interpretation: attachment is relationship-specific and context-sensitive. No category replaces a full account of teaching, learning, communication, health, disability, culture, family, peers and present stress. Use several sources and retain competing explanations.
Better interpretation: adult self-reports ask about current close relationships and usually return category or dimension scores. They are not the Strange Situation or AAI, and they do not recreate an infant classification.
Better interpretation: early relationships matter, and studies find some continuity. They also find change. Measures, relationships and settings change across development. Avoid hardwired or irreversible language.
Better interpretation: match support to observed access, learning and relationship needs. A quiet learner may need explicit help routes; a distressed learner may need a calm pause and clear next step. These responses do not require a style label.
Better interpretation: a study may find an association between relationship quality and cortisol, working memory or behaviour. Report the design and measured outcome. Do not invent a brain pathway, complete protection or “rewiring” that the study did not test.
Attachment theory has had great reach, but its research is not one block of proof. Studies use distinct ages, bonds, tests and results. Groups can hide a full range of scores.
Culture, sample choice and shared causes make the results hard to read. School claims thus need tighter words than most short guides use.
The word “attachment” has more than one use. It can mean an evolved system, a bond, acts under stress, a mental model, a test code or an adult survey score. Two readers may seem to agree while they mean quite different things.
A code is only as sound as the test and training. One group label also drops much of the range within and across patterns. Fraley and Roisman (2014) show why scales need thought in AAI work. Groups can still help, but a type label supports only a narrow claim.
Work across cultures is still in debate. Studies find both shared bond roles and gaps in acts and rates. A small local sample cannot stand for a whole land. Do not rank cultures by security or treat self-rule as one sign of sound growth.
A long-term forecast is still about chance. Later results stem from many linked forces. Dropout from a study, test change and shared methods can skew the size of a link. A modest mean cannot diagnose one learner or promise what comes next.
School use adds one more step. Work on infant-carer bonds does not tell staff which plan to use in class. Studies of teacher-child ties are a closer fit, but the tie is not often first or sole.
Each scheme needs its own proof. Bowlby's name is not enough.
Health terms also need care. An attachment pattern, attachment need and attachment disorder are not the same. NICE (2015) ties diagnosis to set rules and expert checks. A teacher should notice, help, log and refer, not diagnose.
Source strength also varies. Key books and code guides shaped the field but do not answer each new point. A meta-analysis can give a strong review, yet it keeps flaws from the work it pools.
This guide uses sound sources and clear limits. It cannot end each live debate.
These resources keep each attachment claim next to its model, method and limit. Teachers, SENCOs and scholars can use them to check sources, plan staff learning and reject unsafe classroom labels. They are not attachment tests, child profiles, diagnoses or treatment guides.
Bowlby, theory and mechanism: an evolved, goal-led system supports closeness, safety and later play. It does not assign a child to a style from one act in class. Source trail: Bowlby (1969); Rajecki et al. (1978).
Ainsworth, field work and infant test: the Strange Situation uses set spells of play, time apart and reunion. The first work used A, B and C. Later work added D coding. Source trail: Ainsworth et al. (1978).
Main and peers, D coding and adult talk: D is an infant code, not a disorder. The AAI codes how adults speak about past bonds. It is not a romantic-style quiz. Source trail: Main et al. (1985); Granqvist et al. (2017).
Adult social models: adult surveys use groups or scales for close and romantic ties. Their scores do not recreate an infant code or prove a life history. Source trail: Hazan and Shaver (1987); Bartholomew and Horowitz (1991); Fraley et al. (2000).
Five checks: name the model and its proposed process; the test, task, interview or survey; the age and bond assessed; whether the result is a code, score, group link or cause test; and what the result cannot tell us about one child.
School rule: record the task, need, context and response. Do not infer an attachment style. Keep teaching, speech, SEND, health, culture and safeguarding in view.
Two A4 pages for tracing one claim through its model, age, bond, test, sample, result and limits. The second page checks cause, culture, school transfer and other live explanations.
Twelve duplex cards for staff learning, SENCO review or academic study. They correct errors about theory, Ainsworth, D coding, diagnosis, adult measures, stability, culture, cause and classroom support.
Attachment theory explains protection-seeking and exploration within close relationships, while attachment research uses several methods across development. The answers below keep infant classifications, adult measures, classroom observation and attachment disorders separate. They are summaries, not diagnostic guidance.
The main modern tradition is Bowlby's ethological attachment theory, developed alongside Ainsworth's observations. Earlier learning and conditioning accounts proposed different mechanisms. Later work added infant classifications, representation-based methods and adult relationship models. It is more accurate to compare their mechanisms and methods than to call four attachment styles four theories.
No. Ainsworth and colleagues reported three organised Strange Situation classifications: avoidant, secure and resistant or ambivalent. Main and Solomon later developed coding for disorganised or disoriented behaviour. The fourth classification should not be attributed to Ainsworth alone.
No. Disorganised attachment is a research classification coded within the infant Strange Situation. Reactive Attachment Disorder and Disinhibited Social Engagement Disorder are clinical diagnoses with formal criteria. Granqvist et al. (2017) warn against treating the infant classification as a fixed trait, maltreatment test or disorder.
No valid attachment classification follows from ordinary classroom behaviour alone. Teachers can record when difficulty occurs, what demand was present and which response helped. They should consider learning, communication, SEND, health, relationship, culture and safeguarding factors, then use school referral routes when needed.
No. Infant classifications come from structured caregiver observations. The AAI codes discourse about attachment experiences. Adult romantic research often uses self-report categories or anxiety and avoidance dimensions. The traditions are related, but their labels and scores are not interchangeable.
No. Early relationships can matter, and meta-analysis finds some stability over time. Change also occurs, and later measures often assess different relationships or constructs. A first classification is a probabilistic research result, not a permanent identity or forecast.
No. Teacher-child relationship quality is associated with engagement and school outcomes in several studies. The designs, ages and measures vary, and association does not prove one attachment mechanism caused attainment. Predictable, respectful teaching is sound practice without an inflated causal promise.
Ask what assessment produced the label, which relationship and age it concerns, and who completed the assessment. Replace informal labels with observable strengths, barriers, triggers and helpful responses. Keep SEND and health needs visible, protect adjustments and use specialist or safeguarding routes where appropriate.
The references below support the theory, method and inference boundaries used in this guide. Links go to publisher records, DOI records or current guidance. Foundational books are included because later papers do not replace the original constructs.
