Skip to content

AQ-50 Test: Autism Spectrum Quotient Scoring & Accuracy

The AQ-50 Test: Scoring, Accuracy and What It Actually Tells You

Fifty questions. Twenty-five years of research. And a cutoff score that a major clinical study found performed no better than chance.

Items50
TimeAbout 10 minutes
Score range0 to 50
Common cutoff32
Published2001
ForAdults with no intellectual disability

The AQ-50 Test is the most widely used adult autism questionnaire in the world. If you have read a research paper about autistic traits in adults, there is a good chance it used this instrument. If you have taken an online autism quiz, there is an even better chance it was a copy of this one.

That ubiquity is worth understanding properly. It does not mean the AQ-50 Test is the most accurate screener available. It means it arrived early, it was free, and a generation of researchers built on it. Those are different things.

This page covers what the instrument is, how it is scored, who it works for, who it misses, and what the evidence actually shows about its accuracy. Some of that evidence is uncomfortable.

What is the AQ-50 Test?

The Autism-Spectrum Quotient was published in 2001 by Simon Baron-Cohen and colleagues at the Autism Research Centre in Cambridge. The full version contains fifty statements. You rate each one on a four-point agreement scale: definitely agree, slightly agree, slightly disagree, definitely disagree.

Each item scores either one point or zero. There is no partial credit. Agreeing slightly counts the same as agreeing definitely, which is one of the design decisions that has drawn criticism over the years. The total runs from 0 to 50.

The fifty items of the AQ-50 Test were written to cover five areas, ten items each:

  • Social skill. Comfort and competence in social situations.
  • Attention switching. Moving between tasks, and tolerance of interruption.
  • Attention to detail. Noticing patterns, numbers, small changes.
  • Communication. Conversation, turn-taking, reading intent.
  • Imagination. Pretend play in childhood, imagining fictional situations, picturing another person’s perspective.

Roughly half the items are reverse-scored, so agreeing is the autistic-trait response on some and disagreeing is on others. That was deliberate. It makes the questionnaire harder to answer on autopilot than many of its imitators.

The original study

Baron-Cohen’s team gave the instrument to 58 adults with a diagnosis of Asperger syndrome or high-functioning autism, 174 randomly selected controls, 840 Cambridge students, and 16 winners of the UK Mathematics Olympiad. The autistic group scored a mean of 35.8. The controls averaged 16.4. Eighty percent of the autistic group scored 32 or above, against two percent of controls.

That is where the famous cutoff of 32 comes from. It is worth sitting with the design of that study for a moment, because it explains a great deal about what happened next.

How the AQ-50 Test is scored

Scoring the AQ-50 Test is simple, which is part of the appeal. One point for each autistic-trait response. Add them up.

Score Conventional reading
0 to 10 Few autistic traits reported. Below the population average.
11 to 21 Around the typical range. Average for non-autistic adults is roughly 16 or 17.
22 to 25 Above average. Common among people in technical and mathematical fields.
26 to 31 Elevated. Some clinical studies use 26 as a referral threshold.
32 to 50 At or above the original cutoff. Eighty percent of the diagnosed group in the 2001 study scored here.

Those bands get repeated across hundreds of websites as though they settle something. They do not. What follows is the part those sites leave out.

How accurate is the AQ-50 Test, really?

This is the section that matters most, and it is where the honest answer diverges sharply from the popular one.

The 2001 validation compared diagnosed autistic adults against randomly selected controls. That design flatters any instrument. Separating diagnosed autistic people from the general public is an easy task. The hard task, the one that matters clinically, is separating autistic people from other people who have turned up at a clinic because something is wrong.

In 2016, a team at the Maudsley Hospital in London tested exactly that. They looked at 476 adults consecutively referred to a national autism assessment service, 73 percent of whom went on to receive a diagnosis after ADOS and ADI-R assessment.

The result was striking. Self-reported scores did not significantly predict who received a diagnosis. At the conventional cutoff of 32, the instrument performed no better than chance. The lower cutoff of 26 beat chance, but with a specificity of 0.20, meaning it correctly identified only one in five of the people who did not have autism. The authors wrote that national guidance recommending the questionnaire in adult autism assessment “may need to be reconsidered.”

A separate Dutch study in 2021 was more favourable, though still mixed. In 92 adults referred to an outpatient service, an optimal cutoff of 26 gave a positive predictive value of 86 percent, which is good. The negative predictive value was 58 percent. Forty-two percent of the autistic patients were missed entirely. At the conventional cutoff of 32, 59 percent were missed.

Put those together and a consistent picture emerges. A high score on the AQ-50 Test in a clinical population carries real information. A low score carries much less than people assume. If you score 20 and conclude you are not autistic, the evidence does not support that conclusion nearly as strongly as the number feels like it does.

Why the two studies disagree

Mostly because of who was in them. Predictive values are not properties of the AQ-50 Test, or of any questionnaire. They are properties of a questionnaire applied to a particular population. Change the proportion of autistic people in the room and every number moves.

The Maudsley sample was a national tertiary referral centre, so the non-autistic group contained unusually complex cases. The Dutch sample was a general outpatient service. Neither is wrong. They are measuring different things, and anyone quoting a single accuracy figure for this instrument without naming the population is overselling it.

Who the AQ-50 Test is for

The AQ-50 Test was built for and validated on adults of normal intelligence. That is the population where it means something.

It is a reasonable starting point if you are an adult wondering about autism for the first time, if you want a result you can compare against a large published literature, or if you are preparing for a conversation with a doctor and want a number to bring.

It is a poor choice in several situations. If you are under 16, the adult version is not for you. If you have an intellectual disability, the self-report format assumes a level of introspective access the instrument never tested for. And if you have spent your life masking, read the next section carefully.

The masking problem

The AQ-50 Test asks how you are. It does not ask what it costs you to be that way.

Consider the item about whether you find social situations easy. Someone who has spent thirty years building an elaborate set of rules, scripts and rehearsals, and who now passes comfortably in most conversations, can answer honestly that they manage fine. The questionnaire scores that as a non-autistic response. The thirty years of engineering that produced it are invisible to the instrument.

This is the single biggest reason adult screeners miss people, and it falls disproportionately on women and on anyone diagnosed late. It is also precisely what the CAT-Q Test was built to measure. If your score here came out lower than your experience suggests it should, that instrument will tell you more than retaking this one.

Strengths and weaknesses of the AQ-50 Test

Strengths

  • An evidence base spanning more than two decades and hundreds of studies
  • Quick. Around ten minutes.
  • Five subscales give a profile, not just a number
  • Freely available for personal and non-commercial use
  • Reverse-scored items make it harder to game than most online quizzes
  • Translated and validated in many languages

Weaknesses

  • Asks about the present, with no account of your developmental history
  • Underperforms for women and for anyone who masks
  • The binary scoring throws away the difference between slightly and definitely
  • Elevated by anxiety, which can produce false positives
  • The cutoff of 32 has not held up in clinical referral samples
  • Subscale reliability is weaker than the total score

Controversies around the AQ-50 Test

The cutoff

The cutoff used by the AQ-50 Test was derived from one sample in 2001. Thirty-two. It has been treated as settled ever since. The Maudsley study found it performed no better than chance in a referral population, and the Dutch study found it missed 59 percent of autistic patients. Several researchers now argue for 26. Others argue the instrument should not be used for classification at all, only as a dimensional measure of traits.

The anxiety confound

In the Maudsley sample, generalised anxiety disorder was the only significant predictor of a higher short-form score. That is a serious finding. It suggests that some of what the questionnaire detects is not autism but distress, and anxiety is extremely common in people who seek an autism assessment.

Sex and gender bias

The original norms came from a predominantly male diagnosed sample, reflecting who was being diagnosed in 2001. Autistic women tend to score lower than autistic men on self-report instruments while being no less autistic. Research on sex bias in the short form has produced mixed results, but the concern has not been resolved.

The “everyone is a bit autistic” problem

Because the instrument measures traits that are continuously distributed in the population, plenty of non-autistic people score in the twenties. Engineers, mathematicians and physicists score higher on average than humanities students. That is an interesting finding about human variation. It is not evidence that those groups are autistic, and the questionnaire was never designed to carry that weight.

How the AQ-50 Test compares to other instruments

Instrument Items Asks about Best for
AQ-50 Test 50 The present A broad trait profile with a large literature behind it
AQ-10 Test 10 The present A two-minute triage gate, nothing more
RAADS-R Test 80 Your whole life Depth, and a developmental dimension this instrument lacks
RAADS-14 Screen 14 Your whole life Fast screening in psychiatric settings
CAT-Q Test 25 Masking effort Explaining a low score on any of the above
ADOS-2 Test Observation Observed behaviour Formal diagnosis, clinician administered
ADI-R Test 93 Developmental history Formal diagnosis, informant interview
M-CHAT Screening Test 20 Toddler behaviour Children 16 to 30 months, not adults

The most useful comparison is with the lifetime instruments. The AQ-50 Test asks how you are now. Autism is developmental, which means it was there in childhood, and an instrument that never asks about childhood is missing half the question. That is the main structural advantage the RAADS-R Test holds over it.

If you want something broader than autism alone, our Neurodivergent Test covers seventeen conditions across ninety-one questions and reports impact separately from traits. For a shorter autism-specific screen written in plain language, the Free Autism Test takes about six minutes.

Frequently asked questions about the AQ-50 Test

What is a normal AQ score?

The average for non-autistic adults is around 16 or 17. Men average slightly higher than women. Scores in the low twenties are common and unremarkable, particularly among people in technical fields.

I scored 32. Am I autistic?

No. You have reached the threshold the original study used, which means this is worth looking into further. It does not mean you have autism. In a clinical referral population that threshold performed no better than chance, so treat it as a prompt rather than a finding.

I scored below 32 but I am still certain something fits. What now?

This is the most common message people send about this instrument, and the evidence supports taking it seriously. Between 42 and 59 percent of autistic patients in clinical studies scored below the cutoff. A low score here is weak evidence of absence. Try the CAT-Q Test, which measures masking, and a lifetime-format instrument such as the RAADS-R Test.

Can you fail the AQ-50 Test?

There is nothing to fail. It is not a test of ability and there are no right answers. The word test is used loosely here and throughout this site because that is what people search for, but it is a questionnaire about preferences and experiences.

Does the AQ-50 Test diagnose autism?

No, and nothing self-administered does. A diagnosis requires a clinician, a developmental history, and consideration of everything else that could explain your difficulties. Instruments like the ADOS-2 Test and the ADI-R Test exist for that purpose.

How long does it take?

Ten minutes. Less if you answer on instinct, which you should. Constructing scenarios where an item might apply is a good way to produce a meaningless score.

Is the AQ-50 Test free?

The items are freely available for personal and research use. Commercial use requires permission from the Autism Research Centre. You should not have to pay to take it.

Can I use my score to get a diagnosis or an assessment referral?

It can help a conversation along, but do not expect it to do the work. Clinicians know the limitations. What moves a referral forward is concrete examples of difficulty in specific situations, ideally with something from your childhood. Bring those.

Why did I score differently the second time?

Mood, recent events and how you interpret ambiguous items all shift the result. Variation of a few points between sittings is normal. Variation of fifteen points suggests you were answering about different periods of your life, or that one of the sittings was not honest.

Does the AQ-50 Test work for women?

Less well, and that is well documented. Autistic women score lower on average than autistic men while being no less autistic, which is generally attributed to camouflaging and to criteria originally built from studies of boys. If you are a woman with a borderline score, that is a reason to look further rather than to stop.

What about the AQ-10 and the other short versions?

The AQ-10 Test is a ten-item triage version used by the NHS. It is much faster and considerably less informative, and the same Maudsley study found it performed no better than chance at its recommended cutoff. There is also an abridged 28-item version with better subscale structure than the full fifty.

I have ADHD. Will that affect my score?

Probably. Attention switching difficulties appear in both conditions, and anxiety raises scores independently. Separating autism from ADHD is difficult even with full-length instruments administered by clinicians, and a self-report questionnaire cannot do it. Our Neurodivergent Test screens both alongside fifteen other conditions.

What to do with your result

If the AQ-50 Test put you high and the difficulties are costing you something real, write down three or four concrete examples before you speak to anyone. Not “I struggle socially.” Something closer to “I left my last job because the open-plan office was unbearable, and I have turned down two promotions that involved client meetings.”

Specifics move referrals. Scores mostly do not.

If you scored high and nothing is actually costing you anything, that is a legitimate and common result. Traits without impairment are not a disorder. Plenty of people recognise themselves in these descriptions and are doing perfectly well.

If you scored low and still feel certain, do not let a number overrule your own experience. The evidence on low scores is genuinely weak. Look at masking, look at a lifetime-format instrument, and talk to someone.

This page is information, not medical advice. No questionnaire diagnoses autism. If your difficulties are affecting your daily life, your doctor is the right place to start.

Sources