Autism Test

Use the Autism Test

Free autism test using the AQ-10 screening questionnaire: score 10 items against the 6-point cut-off and see what a positive result really means for you.

How to Use Autism Test

1. Step 1: Answer all ten AQ-10 statements

   Each statement offers four options: definitely agree, slightly agree, slightly disagree, definitely disagree. Answer for how you usually are across your whole life, not how the last fortnight has gone. There is no neutral option on purpose, so pick the side you lean toward even when both feel partly true.

2. Step 2: Watch the scoring-direction badge on each item

   Every item card is tagged either agreeing scores 1 or disagreeing scores 1. Items 1, 7, 8 and 10 score for agreement; items 2, 3, 4, 5, 6 and 9 score for disagreement. Each option also shows whether it is worth 1 point or 0, so you can see the point land as you answer.

3. Step 3: Read your total against the cut-off of 6

   The sticky bar tracks your running total out of 10 with the cut-off marked on the progress bar. A score of 6 or more is the point at which NICE guideline CG142 suggests offering a full diagnostic assessment. The result card also tells you how many points above or below the line you landed.

4. Step 4: Set your base rate in the probability panel

   Pick the group you belonged to before you took the test: a randomly chosen adult at 2%, an adult who already suspects they are autistic at 30%, or a specialist clinic referral at 50%. Choose Set my own base rate to move the slider yourself. The panel converts your result into a post-test probability and draws it as 100 people.

5. Step 5: Stress-test the accuracy assumptions

   Open the test-accuracy panel and edit sensitivity and specificity. They default to the 88% and 91% reported in the AQ-10 development study, which used a sample already enriched for autism. Drop specificity to 60% and watch the false-alarm column in the per-1,000 table grow.

6. Step 6: Copy the item-by-item summary

   Copy summary saves every answer, the point it scored, your total, the cluster split, and the probability assumptions you used. Bring that to a GP appointment rather than the score alone, because which items you endorsed matters more to an assessor than the number.

Key Features

• Full adult AQ-10 with the correct scoring direction on every item — six of the ten score a point for disagreeing
• Post-test probability panel that turns your score into a real chance of diagnosis at any base rate
• Natural-frequency table and 100-person chart separating true positives from false alarms
• Editable sensitivity and specificity so you can stress-test the 88% / 91% development-sample figures
• Item-cluster split across social insight, attention to detail and attention switching
• Copy or print an item-by-item summary to bring to a referral appointment

Understanding Results

Formula

The AQ-10 total is a plain count with one asymmetry that trips up most hand-scoring. Each of the ten items offers four options — definitely agree, slightly agree, slightly disagree, definitely disagree — and scores 1 point if the answer falls on the autism-consistent side, 0 if it does not. Strength does not matter: “slightly” and “definitely” are worth exactly the same. The range is 0 to 10.

The asymmetry is direction. Only items 1, 7, 8 and 10 score for agreeing. Items 2, 3, 4, 5, 6 and 9are worded the other way round — “I find it easy to read between the lines,” “I know how to tell if someone listening to me is getting bored” — so they score for disagreeing. Six of the ten items are reverse-keyed, which is why a form filled in without checking direction can be wrong by several points in either direction.

The probability panel runs a separate calculation on top of the total: Bayes’ rule. With sensitivity Se, specificity Sp, and a pre-test probability p, the chance you are autistic given a positive screen is Se × p / (Se × p + (1 − Sp) × (1 − p)). The likelihood ratios — LR+ = Se / (1 − Sp) and LR− = (1 − Se) / Sp — are the same arithmetic expressed in a form that does not depend on p.

Reference Ranges & Interpretation

There is exactly one published threshold: 6 or more out of 10. NICE guideline CG142 on autism in adults names the AQ-10 as the screening tool to use and treats a score at or above 6 as the trigger for offering a comprehensive diagnostic assessment. Below 6, the questionnaire alone does not prompt referral. Bands of 0–3, 4–5, 6–7 and 8–10 appear in the results table here for orientation only — they are our reading, not published cut-points, because the instrument has a single official line.

The development study by Allison, Auyeung and Baron-Cohen reported 88% sensitivity and 91% specificityat that cut-off. Those two numbers alone are not an answer to “am I autistic,” and this is the part most autism test pages skip. At an adult population base rate near 2%, screening 1,000 people yields about 106 positive screens of which only 18 are autistic — a positive predictive value near 17%. Among adults who already suspect it strongly enough to seek assessment, take a pre-test probability of 30% and the same score of 7 carries roughly an 81%chance. Identical answers, identical total, five-fold difference in meaning. The likelihood ratios stay fixed at LR+ 9.8 and LR− 0.13.

Assumptions & Limitations

The 88% / 91% figures come from a sample already enriched for autism, and independent evaluations in less selected groups have reported materially lower specificity. That is why sensitivity and specificity are editable here rather than hard-coded: set specificity to 60% at a 30% base rate and the positive predictive value falls from 81% to about 45%. Any online autism test that reports a verdict without telling you its assumed base rate is reporting half a calculation.

Several items ask whether you find something easyrather than whether it costs you effort. An adult who has spent thirty years consciously learning to read faces will often answer “yes, easy,” score 0, and walk away with a total that understates the picture. This masking effect is thought to fall disproportionately on autistic women and on late-diagnosed adults generally; the diagnosed sex ratio of roughly 4:1 male to female is wider than the true ratio, which a 2017 meta-analysis placed nearer 3:1. A low score in someone with a lifetime of unexplained social exhaustion is weak evidence, not a clearance.

Three further boundaries. The AQ-10 was validated for age 16 and overas self-report; the 4–11 and 12–15 versions use different items with a parent as informant, so they are not interchangeable. It cannot distinguish autism from conditions that produce overlapping answers, including ADHD, social anxiety disorder and the aftermath of developmental trauma — separating those is the substance of a real assessment, not a questionnaire. And it measures traits, not impairment: a DSM-5 diagnosis additionally requires that the traits limit functioning and were present from early development. This tool does not provide a diagnosis or medical advice.

Complete Guide: Autism Test

The autism test that decides who gets referred for a full assessment in the NHS is ten questions long, takes about two minutes, and has exactly one number attached to it: 6. That instrument is the AQ-10, a short form of the Autism Spectrum Quotient published by Allison, Auyeung and Baron-Cohen in 2012 and named in NICE guideline CG142 as the screening tool for adults who may be autistic. Score 6 or more and the guideline says you should be offered a comprehensive diagnostic assessment. Score 5 and it does not. This guide covers how the ten items are actually scored — six of them backwards — what the cut-off does and does not mean, and the piece nearly every autism screening page leaves out: what your score is worth once you account for who you were before you took it.

Ten Items Lifted From a Fifty-Item Questionnaire

The parent instrument, the 50-item AQ, dates from 2001 and measures autistic traits across five domains: social skill, attention switching, attention to detail, communication and imagination. Fifty items is fine for a research protocol and useless in a ten-minute GP appointment, so the 2012 study asked a narrower question — which items carry the discriminative weight? Ten survived, drawn from across those domains, and the short form kept the same four-option response scale.

Notice what the response scale does not include: a neutral option. You get definitely agree, slightly agree, slightly disagree, definitely disagree, and nothing in the middle. That is deliberate. A midpoint on a ten-item instrument would collect every ambiguous answer and flatten the score toward the middle, which is exactly where a screening cut-off cannot afford noise. The cost is that you have to commit on items where both sides feel partly true, and that discomfort is a real source of scoring variance rather than a flaw you should try to avoid.

One structural point worth knowing before you read your result: the AQ-10 is validated as a single total. Its 50-item parent has five subscales with their own scores; the short form does not. When our calculator splits your points into social insight, attention to detail and attention switching, that split is descriptive — useful for knowing which items to talk about at an appointment, not a second result with its own threshold.

Six of the Ten Score a Point for Disagreeing

This is the mechanical detail that produces most wrong totals. One point per item, ten items, maximum 10 — but the point is awarded for the answer on the autism-consistent side, and for six of the ten items that side is disagreement. Item 5 reads “I find it easy to read between the lines when someone is talking to me.” Agreeing with that is the non-autistic answer and scores nothing. Strength is irrelevant throughout: slightly disagree and definitely disagree are both worth exactly 1 point.

Four agree-keyed items, six disagree-keyed. If you hand-score by counting how many statements “sound like you” without checking direction, you can be wrong by up to 10 points on a 10-point scale. The calculator above tags every item with its direction and marks each option 1 point or 0 before you pick, which removes the error class entirely.

What Crossing 6 Actually Buys You

A cut-off is an administrative decision, not a fact about you. NICE set 6 as the point at which the expected benefit of assessing someone exceeds the expected cost of assessing them unnecessarily — a judgement about clinic capacity and waiting lists as much as about autism. Cross it and the guideline says a comprehensive assessment should be offered. That is the whole content of the threshold. It does not say you are autistic, and it does not say a 6 is meaningfully different from a 5.

Five is the genuinely awkward score. One item re-read on a different day — item 3, say, on whether you find it easy to do more than one thing at once, which depends heavily on what you were doing that week — and you are at 6. Test-retest movement of a point or two on a ten-item instrument is ordinary measurement behaviour, not a change in you. Anyone landing at 4 or 5 who has a long history of the difficulties the items describe should treat the number as a near-miss rather than a verdict.

The development study reported 88% sensitivity and 91% specificity at the cut-off of 6. Read the second figure carefully. Specificity of 91% means 9 in every 100 non-autistic people screen positive. In a room of a hundred adults with no autism at all, nine would cross the line. Whether that matters to you depends entirely on the next section.

The Same Score of 7 Means 17% or 81%

Here is the thing almost no autism test page will tell you: a screening score, on its own, is not a probability of anything. It only becomes one when you combine it with how likely you were to be autistic beforeyou took the test. Two people can produce the identical set of ten answers and the identical total of 7, and the honest answer to “what are the odds?” differs between them by a factor of five.

The arithmetic is Bayes’ rule and it is short. Sensitivity 88%, specificity 91%. Screen 1,000 people whose true autism rate is 2% — roughly the adult population figure — and you get 20 autistic people, 18 of whom the test catches, plus 88 false alarms among the 980 who are not autistic. That is 106 positive screens holding 18 real cases. Positive predictive value: 17%. Now screen 1,000 people who already suspected it strongly enough to go looking for a test, where the true rate might be 30%, and the same instrument produces 264 true positives against 63 false alarms. Positive predictive value: 81%.

Two Readings, and a Caveat on Those Accuracy Numbers

Two practical readings fall out of that table. First, if you found this page because something in your life stopped making sense and autism was the explanation that fit, you are not the 2% row — the act of searching has already moved you down the table, and a positive screen carries real weight. Second, the false-alarm column is driven by the base rate, not by the questionnaire: the test performs identically in every row and the answer still swings from 17% to 91%. That is why the calculator makes you choose a base rate before it will show you a probability.

One caveat on the accuracy figures themselves. The 88% / 91% pair comes from the development sample, which was enriched for autism; independent evaluations in less selected populations have reported lower specificity, and lower specificity hurts fast. Set specificity to 60% at a 30% base rate and the positive predictive value collapses from 81% to roughly 45%. The sensitivity and specificity fields in the calculator are editable for precisely this reason — run your own worst case rather than trusting one paper’s best case.

Camouflaging: Where the AQ-10 Loses People

Look again at items 5, 6 and 9. Each asks whether you find something easy: reading between the lines, telling when a listener is bored, reading a face. None asks what it costs you. An adult who worked out at fourteen that you can infer boredom from a phone check and a shortening reply, and who has run that rule consciously in every conversation since, will answer “yes, I can tell” and score zero on item 6. The answer is truthful. The item still missed them.

This is camouflaging, and it is the best-documented reason the AQ family under-detects. It falls unevenly: the diagnosed sex ratio of about four males to one female is wider than the true ratio, which a 2017 meta-analysis of prevalence studies placed closer to 3:1once ascertainment bias was accounted for — implying a substantial population of autistic women who never reach diagnosis. The CAT-Q was developed specifically to measure camouflaging behaviour, and it is the sensible second questionnaire when an AQ-10 comes out lower than a lifetime of experience suggests it should.

The cost of that constant translation usually shows up somewhere else, which is why so many adults arrive at autism through a mental health route rather than a developmental one. A 2019 meta-analysis of autistic adults found a current anxiety disorder in about 27% and a lifetime rate near 42%, with depression at roughly 23% current and 37% lifetime — several times the general-population figures. If your AQ-10 sits at 6 or above, running the GAD-7 anxiety screener and the PHQ-9 depression screener gives an assessor a fuller picture than the autism score alone, and the PHQ-4 combined checkcovers both in four questions if you want the short version. Sleep is the other common casualty — insomnia is far more common in autistic adults than in the general population, and the Insomnia Severity Index quantifies it on a scale a clinician will recognise.

Four Ways People Score an Autism Test Wrong

• Ignoring reverse-keying.Six of the ten items score for disagreeing. Counting statements that “sound like me” produces a number with no relationship to the published cut-off.
• Answering for this month. The AQ-10 asks about stable traits. A burnt-out fortnight will inflate items 3 and 4 on attention switching, which is a state effect, not a trait. Answer for your life, not your week.
• Reading “easy” as “possible.”Items 5, 6 and 9 ask whether something comes easily. If you can do it but it takes deliberate effort and leaves you drained, that is not “easy” — and answering as though it were is the mechanism described in the section above.
• Reading the total as a probability.A 7 is not “70% autistic.” It is a positive screen whose meaning ranges from 17% to 91% depending on the base rate, as the table above sets out. The score is an input to a probability, never the probability itself.

You Scored 6 or More. What Happens Next?

In the UK the route is a GP appointment and a request for referral to an adult autism diagnostic service; the AQ-10 exists largely to make that conversation concrete. Elsewhere it is usually a clinical psychologist or psychiatrist with adult autism experience. Either way, do not lead with the number. An assessor is not going to weight a self-administered ten-item screen heavily, and the useful material is underneath it: which items you endorsed, concrete childhood examples, school reports if any survive, and an account from someone who knew you before you were ten.

Run the ADHD question at the same time rather than afterwards. DSM-IV explicitly forbade diagnosing ADHD alongside autism; DSM-5 removed that prohibition in 2013, and the two now co-occur at rates far above chance. Assessment services are frequently separate, separately queued, and separately waited on, so arriving with both screeners already completed can save a second year in a queue. The ASRS adult ADHD test is the 18-item WHO instrument, and it takes about three minutes. If your AQ-10 items clustered on attention switching rather than social insight, it is the more informative of the two to run first.

Set expectations on time. Adult autism assessment waiting lists in many NHS areas run beyond a year, and the assessment itself is typically several hours across more than one appointment, frequently using the ADOS-2 or ADI-R alongside a developmental history. The delay is worth raising at the first appointment rather than the third, because in some regions a right-to-choose referral shortens it substantially.

And if you scored 5 or below but the description still fits, the useful move is not to retake the test hoping for a different number. It is to write down the specific situations that go wrong — the meetings that exhaust you, the plans that change and cost you the rest of the day, the sounds nobody else notices — and take those. A questionnaire that gets 12 in every 100 autistic people wrong is not the final word on whether you are one of them.

References

1. Allison C, Auyeung B, Baron-Cohen S. Toward brief “red flags” for autism screening: the short Autism Spectrum Quotient and the short Quantitative Checklist in 1,000 cases and 3,000 controls. Journal of the American Academy of Child & Adolescent Psychiatry, 2012;51(2):202–212.
2. National Institute for Health and Care Excellence. Autism spectrum disorder in adults: diagnosis and management (CG142). NICE, 2012 (updated 2021).
3. Loomes R, Hull L, Mandy WPL. What is the male-to-female ratio in autism spectrum disorder? A systematic review and meta-analysis. Journal of the American Academy of Child & Adolescent Psychiatry, 2017;56(6):466–474.
4. Hollocks MJ, Lerh JW, Magiati I, Meiser-Stedman R, Brugha TS. Anxiety and depression in adults with autism spectrum disorder: a systematic review and meta-analysis. Psychological Medicine, 2019;49(4):559–572.
5. Centers for Disease Control and Prevention. Autism Spectrum Disorder: Data and Statistics. CDC, U.S. Department of Health and Human Services.

Frequently Asked Questions