Almost every adult who has spent time on the internet has at some point taken an attachment-style quiz, scored "anxious" or "avoidant" or "secure," and either nodded slowly or shrugged and closed the tab. The next obvious question — does the result mean anything — is asked surprisingly rarely. So this piece is the answer. There are real, validated attachment instruments. Most online quizzes are not them. And the gap between the two matters, because adults make actual decisions about themselves and their relationships based on a result that may have been generated by twelve poorly-worded multiple-choice items.
The short version: the proper instruments — the Experiences in Close Relationships scale (ECR), its revised version (ECR-R), the Adult Attachment Scale (AAS), and the Adult Attachment Interview (AAI) — are reasonably accurate measures of the construct they claim to measure, with decades of validation behind them. Most online quizzes are loose copies of these instruments, sometimes with permission and proper translation, often without, and almost always with the dimensional structure flattened into a four-box category result. The flattening is where the accuracy goes.
The Three Generations of Attachment Measurement
Adult attachment measurement has gone through three serious generations. The first was John Bowlby's clinical observation in the 1950s and Mary Ainsworth's Strange Situation procedure in 1969–1978, which classified infants into secure, avoidant, and anxious-ambivalent categories based on behaviour at reunion with a caregiver. That work was about babies. It was empirical and observational, not a self-report instrument.
The second generation was the bridge to adulthood. Cindy Hazan and Phillip Shaver in 1987 produced the original three-paragraph self-report — adults read three descriptions of how they felt in close relationships and chose the one that fit best. About 56% picked secure, 25% avoidant, 19% anxious. The instrument was famously rough — three statements is not measurement, it's selection — but it was the foundational bridge that made adult attachment research possible. Kim Bartholomew and Leonard Horowitz in 1991 refined this into a four-paragraph version (Relationships Questionnaire, RQ) that added the fearful-avoidant category and produced a 2×2 model crossed on self-view and other-view.
The third generation is where the proper instruments live. Kelly Brennan, Catherine Clark and Phillip Shaver in 1998 published the Experiences in Close Relationships scale — 36 items measuring two continuous dimensions, attachment anxiety and attachment avoidance. R. Chris Fraley, Niels Waller and Brennan in 2000 produced the revised ECR-R, with 36 items selected via item-response theory for better measurement across the latent trait range. The two dimensions are scored as continuous values, not as categories. Secure, anxious, avoidant and disorganised are convenient labels for regions of the two-dimensional space, not types in the strict sense. (See attachment styles — complete guide.)
The clinical gold standard is separate. Mary Main's Adult Attachment Interview is a structured interview about the participant's childhood relationships, coded for narrative coherence rather than content. The AAI takes roughly an hour, requires a trained coder, and measures the participant's "state of mind with respect to attachment" through how they talk about it rather than what they say. The AAI is the gold standard but it is not a quiz; it cannot be self-administered online.
What Makes an Attachment Test Actually Accurate
Three properties separate good attachment measurement from bad. The first is dimensional rather than categorical scoring. The two dimensions — anxiety (concern with rejection, hyperactivation of the attachment system) and avoidance (discomfort with closeness, deactivation of the attachment system) — are continuous. Most adults are not strongly in one corner; they're in the middle, leaning. A categorical four-box result discards almost all of this information.
The second is internal consistency and test-retest reliability. The ECR-R has Cronbach's alpha typically above .90 on both subscales — meaning the items hang together — and test-retest reliability around .50 to .70 over months to years. The original three-paragraph Hazan and Shaver instrument has lower test-retest reliability (around .50 at three weeks in some studies), partly because asking someone to pick one paragraph forces an arbitrary choice when the real result is two-dimensional.
The third is convergent validity. A good attachment measure should correlate appropriately with related constructs — relationship satisfaction, interpersonal trust, emotion regulation strategies — and predict outcomes in the literature. The ECR-R does. The 12-item Buzzfeed-tier quiz almost certainly does not, because nobody has measured.
The Big Online-Quiz Tells
Tell 1 — Categorical result, no dimensions
If the quiz tells you "you are 80% anxious-preoccupied!" with no separate avoidance score, it has flattened a two-dimensional construct into a single label. The real result is two numbers, not one — your anxiety score and your avoidance score. A test that gives you only a category has thrown away half the information.
Tell 2 — Fewer than 18 items
The validated ECR-R has 36 items, 18 per dimension. Fraley's online "short ECR-R" uses around 9 items per dimension, which is a defensible compromise. A 12-question total quiz cannot measure two independent dimensions reliably. The arithmetic does not work — too few items, too much noise, especially for adults whose scores are near the boundaries.
Tell 3 — No statement of which scale it's based on
The proper quizzes cite ECR or ECR-R or RQ or AAS. The Buzzfeed-tier ones don't, because they're loosely-paraphrased item collections that haven't been validated. Look for a citation. If there isn't one, the result is closer to a horoscope than a measurement.
Tell 4 — Romance-context-only or romance-context-absent
Adult attachment can be measured in romantic relationships, in close friendships, or as a generalised disposition. The good instruments specify which they're measuring (the ECR specifies "close relationships" generally; the AAS is romance-specific; the AAI is parental-childhood-focused). Quizzes that switch frames mid-quiz — half items about partners, half about parents, half about friends — produce muddled scores.
What Online Quizzes Are Useful For
Even with the limitations, an online attachment-style quiz can be useful for one thing: it gives you a rough working hypothesis. If a properly-built short quiz (eg. Fraley's online ECR-R, which is academic, free, and produces dimensional scores) lands you firmly in the anxious-leaning quadrant, that's a defensible starting point for self-investigation. The result is not a diagnosis. It is a hypothesis to test against your behaviour across several relationships.
The test you actually run is months long: do the patterns described under "anxious attachment in adulthood" — preoccupation with the partner's availability, catastrophising silence, over-monitoring tone — actually show up in your relationships? If yes, the quiz hypothesis is supported. If no, it isn't. The behavioural data outweighs the questionnaire data. (See anxious attachment in dating — deep guide.)
The Seven Tests People Most Often Find Online — and How They Stack Up
Without naming and shaming specific websites, the seven categories of online attachment test break down as follows:
Category 1: Fraley's online ECR-R at yourpersonality.net. Academic, free, used in research. Produces two dimensional scores. The closest thing to a properly-validated free online test. The result is conservative and dimensional — it tells you where you sit on the two axes, not which label you are.
Category 2: The original Hazan and Shaver three-paragraph test. Often appears on relationship blogs. Famously rough but historically influential. Useful only as an introduction to the concept. The three-paragraph forced choice flattens the dimensions further than even a categorical four-box result.
Category 3: The four-paragraph Bartholomew & Horowitz RQ. Better than three. Still single-item per category. Useful for context, not for accurate placement.
Category 4: Branded therapy-website quizzes (eg. those linked to attachment-style coaching products). Mixed quality. Some are loose ECR-R translations; some are bespoke and unvalidated. Almost always categorical, almost always commercially incentivised to produce a clear result. Read the small print about what scale they say they're based on.
Category 5: Quizzes embedded in popular books. Quality varies dramatically. The Levine & Heller Attached quiz is loosely based on the AAS and produces a category result; useful as a starting point, not authoritative. Other book quizzes are weaker.
Category 6: Buzzfeed-style "what attachment style are you?" quizzes with cartoon images and ten items. These produce entertainment, not measurement. The result correlates more strongly with which images you found cute than with your attachment dimensions.
Category 7: AI-chatbot attachment assessments. A newer category. The chatbot asks open-ended questions and produces a category result. The quality varies with the underlying prompt. Even when the prompt is good, the validation is non-existent — these have not been compared to ECR-R scores in published research. Use cautiously.
For dimensional accuracy, Fraley's online ECR-R is the available best option. Take it once carefully when you're not in an unusually emotional state, and trust the dimensional output more than the category label. (See attachment style quiz — evidence-based.)
Attachment is one input. Compatibility is four.
LoveCertain weights attachment style at 20% of compatibility — alongside values (40%), life stage (25%) and communication style (15%). Only matches above 70%. £49 once. Full refund if no relationship in 90 days. £99 bonus if there is.
Why Your Score Can Move Between Quizzes
It is normal to take three attachment quizzes in a week and get three different results. There are several reasons. First, the quizzes themselves measure different things — one is biased toward romantic context, another toward parental context, a third toward general close relationships. The construct varies subtly by frame.
Second, your state on the day affects self-report. An adult who has just had a difficult conversation with a partner will report higher anxiety scores than they would on a calm Sunday. The trait is steady; the day-state moves it ±10–15% on a 7-point scale. Take the test on different days and the noise will average out. (See attachment style changes over time for the trait vs state distinction.)
Third, the categorical cutoffs are not standardised. One quiz puts the secure cutoff at anxiety < 3.5; another at anxiety < 4.0. Adults whose true score is around 3.7 will be "secure" on one and "anxious" on the other. The dimensional reality hasn't moved; the label flipped because the line moved.
The Honest Limits of Self-Report
Even the proper instruments share limits. Self-report attachment measures correlate only modestly (around r ≈ .15 to .30 in Roisman and Fraley's meta-analytic work) with AAI-coded attachment based on narrative coherence. That is — what adults report on questionnaires about how they feel in close relationships is not the same construct as what their narrative about childhood reveals to a trained coder. Both are real attachment measures; they measure different facets.
For the purposes of dating and adult relationship work, the self-report (ECR-R style) is the more directly relevant measure. It captures how you experience your current and recent close relationships. The AAI captures something deeper about your representation of attachment-relevant experiences. Both are useful; they are not interchangeable. (See attachment theory in dating.)
What to Actually Do With Your Result
Step 1 — Treat the result as a hypothesis, not a verdict
Your quiz score is the start of an investigation, not its conclusion. The honest framing: "the quiz suggests I lean anxious-preoccupied — let me observe whether the patterns it describes actually show up in my recent relationships." Then you watch your behaviour for a month with the hypothesis in mind. Confirmed in your behaviour? Useful. Not confirmed? The quiz was probably noise.
Step 2 — Take the test twice, weeks apart
One result from one quiz on one day is not data. Two results from the same instrument three weeks apart, with both yielding similar dimensional scores, is data. If your scores swing dramatically across short intervals on the same instrument, the result is unreliable for you and you should reduce your weighting of it. State-day variation is real and worth filtering out.
Step 3 — Read the dimensional score, not the label
"You are anxious-preoccupied" is the label. The dimensional score might be 4.2 anxiety / 2.8 avoidance, which is mildly anxious-leaning, mildly secure on avoidance. The label suggests something stronger than the underlying score warrants. Always look for the numbers behind the label and use them, not the category, when reasoning about yourself. (See secure attachment traits — full list.)
Step 4 — Pair the quiz with behavioural observation
The quiz is one input. Observation of your patterns across two or three relationships is the more authoritative one. Are you the partner who checks the phone twice an hour, or are you the partner who lets it sit for three days? Are you the partner who needs the post-conflict conversation that evening, or the partner who needs 48 hours? The behaviour is the validation.
The 30-day attachment audit
Week 1: take Fraley's online ECR-R at yourpersonality.net. Note the two dimensional scores. Week 2: write three short paragraphs describing how you typically respond when (a) a partner is slow to reply, (b) a partner seeks reassurance, (c) a partner is emotionally distant for a day. Week 3: ask a long-term friend to read your paragraphs and tell you whether they recognise the patterns. Week 4: re-take the ECR-R. If the scores match week 1 and the friend confirmed the behavioural patterns, you have a real working hypothesis about your attachment dimensions.
The wider research
R. Chris Fraley's published work on attachment measurement is the single most useful body of scholarship for understanding how these instruments work. Mikulincer and Shaver's Attachment in Adulthood (Guilford, 2nd ed. 2016) compiles the empirical landscape. Glenn Roisman's longitudinal work on the AAI vs self-report distinction is the place to look if you want to understand why your AAI score and your ECR-R score might not match. The field is mature; the popular discourse is, charitably, a generation behind it.
The Certain Letter
Weekly relationship-science briefings. 4-minute read.
Why This Matters For Dating
Two practical implications. First: if you are weighing attachment compatibility with a partner, do not over-weight the result of a single 12-item quiz either of you took on a Tuesday. The variance is too high. Use the result as a conversation prompt, not a decision input. If your respective quiz results say "anxious-avoidant pairing" but your actual relationship shows little of the anxious-avoidant trap pattern, your relationship is the better data than the quiz. (See the anxious-avoidant trap.)
Second: a quiz result of "disorganised" is the one to treat most cautiously. Disorganised attachment is a low-base-rate adult style with serious clinical undertones, and a 10-item online quiz has roughly no chance of distinguishing genuine disorganised attachment from a state-day-anxious response. If a quiz returns "disorganised" and the description resonates, treat it as a prompt to seek professional assessment rather than a self-diagnosis. (See disorganised attachment — honest guide.)
For a primary-source academic overview of the ECR-R and the measurement field, see R. Chris Fraley's attachment-style measurement summary on Simply Psychology, which links through to the original academic instruments.
The Encouragement
Quizzes are not the enemy. They are a low-cost entry point into a useful body of research. The problem is treating their output as identity rather than hypothesis. Read your score lightly. Test it against your behaviour. Use the dimensional numbers if you can get them. And remember the construct itself is dimensional and dynamic, not a label you'll wear for life. Attachment patterns shift with consistent secure-base experiences across years — the score you get today is your working position, not your destiny. (See becoming securely attached as an adult.)
