You also get the figure most free tests leave out: the margin of error. A 16-item test with a published internal consistency near .81 carries a 95% confidence band of roughly ±13 IQ points. A supervised WAIS-IV carries ±4. Both numbers are calculated below rather than asserted, and the distance between them decides how much weight your result can reasonably carry.
Contents · 17 sections
- What you get
- Questions and time
- The four domains
- Why sub-scores are weak
- Raw score to IQ
- Percentile table
- Compared with the WAIS-IV
- How wrong it can be
- What moves your score
- Retaking the test
- What it predicts
- Mensa
- A century of rising scores
- When it is the wrong tool
- What happens to your answers
- Questions people ask
- Sources
What the result screen gives you
The result is four things: a full-scale IQ estimate, a percentile rank, four domain scores, and a confidence band around the total. Read the band before the number. A single score samples your performance on one afternoon, and the band describes how much that sample could differ from your typical performance — which is how clinical reports present IQ and how almost no free test does.
Nothing is emailed to you, and nothing sits behind a payment step. A test that computes your score and then asks for money to display it is running a sales funnel with psychometrics as bait. Our editorial position lists the five things this site has decided never to publish, and a gated result is one of them.
How many questions the test has and how long it takes
Sixteen items, four from each reasoning domain, six answer options apiece, and no countdown clock on any screen. Most people finish in about ten minutes; three-dimensional rotation items are the slowest for many test-takers, which is exactly why the set is untimed.
That choice follows the guidance of the researchers who built the item pool. Their documentation notes that unsupervised online testing has to allow for interruptions and device problems, so a time limit ends up scoring your connection rather than your reasoning. Speed does carry real information about cognitive ability — in a browser, it also carries information about your Wi-Fi, and the two cannot be separated.
One practical consequence of six answer options: pure guessing produces about 2.7 correct answers out of 16. A score built mostly on guesses lands near the floor of the scale, not in the middle of it.
The four reasoning domains behind your score
Four item types feed the total, and each targets a different operation:
Letter and number series
A short sequence of digits or letters, and you identify what comes next.
Matrix reasoning
A three-by-three grid of geometric shapes with one cell missing, in the style of Raven’s Progressive Matrices.
Verbal reasoning
General-knowledge and logic questions that draw on vocabulary and inference.
Three-dimensional rotation
Cubes carrying a different image on each face, where you judge which option shows the same cube turned in space.
The common claim that matrix reasoning is the purest available marker of fluid intelligence turns out to be contested. When the 16-item set was validated against the WAIS-IV in adults, only letter and number series aligned strongly with fluid reasoning (Gf, r = .70); matrix reasoning, verbal reasoning and three-dimensional rotation all correlated most strongly with visual-spatial ability (Gv, r = .35–.75), a result the authors themselves flagged as surprising and inconsistent with CHC theory (Young & Keith, Journal of Psychoeducational Assessment, 2020). A 2025 validation of a children’s version of the same pool reached the opposite conclusion for matrix reasoning, finding it loaded most strongly on Gf as theory predicts (Behavior Research Methods, 2025).
Two well-conducted studies, two different answers, different age groups. Treat your domain profile as a description of how you performed on these specific tasks rather than as a map of your cognitive architecture.
Why your domain scores are less trustworthy than your total
Each domain score rests on four items, and four items cannot support a confident conclusion about anything. One lucky guess moves a domain score by 25%. One misread instruction does the same in the other direction.
Reliability rises with test length, which is why the total across all 16 items is the number worth reading and the four sub-scores are a hint at most. If your rotation score sits two points above your verbal score, that gap is well inside what random variation produces. The same caution applies to any short test that hands you a detailed profile: the more granular the breakdown, the fewer items sit behind each part of it.
How a raw score becomes an IQ number
Your raw score — the count of correct answers — is placed on a normal distribution with a mean of 100 and a standard deviation of 15. IQ is a rank, not a quantity. Nobody possesses 118 units of intelligence; a score of 118 says that this performance beat roughly 88% of the reference sample on this particular set of tasks.
That is also why IQ scores cannot be averaged across different tests or compared across scales without conversion, and why the same person’s number moves when the reference group changes. The build and scoring notes set out the item parameters, the model behind the conversion, and the things this instrument has not established.
What your score means as a percentile
| IQ score | Percentile | Roughly how common |
|---|---|---|
| 130 | 98th | 1 in 44 |
| 125 | 95th | 1 in 21 |
| 120 | 91st | 1 in 11 |
| 115 | 84th | 1 in 6 |
| 110 | 75th | 1 in 4 |
| 100 | 50th | 1 in 2 |
| 90 | 25th | 1 in 4 |
| 85 | 16th | 1 in 6 |
| 70 | 2nd | 1 in 44 |
Two thirds of all scores fall between 85 and 115, and about 95% fall between 70 and 130. Notice how unevenly the scale is spaced: the ten points from 100 to 110 cover 25 percentile ranks, while the ten points from 130 to 140 cover barely two. Differences at the top of the distribution are far smaller in human terms than they look on paper. Our guide to reading a score takes that further, including why percentile is not what most people assume it is.
How accurate a free online IQ test is compared with the WAIS-IV
The strongest evidence for a short public-domain test comes from a head-to-head study with the clinical gold standard: total scores on the 16-item set correlated .81 with WAIS-IV Full-Scale IQ, and the latent general factors of the two instruments correlated .94 (Young & Keith, 2020). The original development paper reported a correlation of .75 with Raven’s Advanced Progressive Matrices after correction for range restriction (Condon & Revelle, Intelligence, 43, 2014).
Those numbers deserve their caveats stated plainly. The WAIS-IV comparison used 97 participants — 67 university volunteers plus 30 people tested at a university assessment centre — and the correlations were corrected for range restriction and unreliability, so they describe a statistical relationship between constructs rather than raw agreement between two score reports. The authors called for replication in larger samples, and no such replication has been published.
An honest summary: a short test built on published, validated items tracks a clinical battery well enough to be informative and not well enough to be decisive. A psychologist administering the WAIS-IV spends 60 to 90 minutes one-on-one, observes how you approach each problem, and interprets the result against your history. A browser does none of that.
How many points your score can be wrong by
A short online test with a reliability coefficient of .81 has a 95% confidence interval of about ±13 IQ points, against roughly ±4 for the WAIS-IV. The arithmetic is standard: the standard error of measurement equals the scale’s standard deviation multiplied by the square root of one minus the test’s reliability, and the 95% interval is 1.96 times that error.
| Test reliability | Standard error | 95% confidence interval |
|---|---|---|
| .98 — WAIS-IV Full-Scale IQ | 2.1 points | ±4 |
| .90 | 4.7 points | ±9 |
| .81 — published for the 16-item set | 6.5 points | ±13 |
| .70 | 8.2 points | ±16 |
A result of 112 here means your true score most likely sits between 99 and 125. That range spans “ordinary” and “well above average”, which is why a two-point gap between you and a friend carries no information at all.
Reviewing the free IQ tests that rank for this query in July 2026, I could not find one that publishes a reliability coefficient or a confidence interval for its own scale. Several quote the WAIS-IV error of ±4 alongside their own results, which is the precision of a different instrument entirely. Before trusting any test’s number, ask what its reliability is — a test that cannot answer has not been calibrated.
What moves your score on the day you take it
Test conditions shift results by several points in both directions, which is large relative to the differences people actually care about. Sleep and freedom from interruption work in your favour. Screen size works against you on rotation items, where a phone forces more of the figure into working memory than a laptop would. Verbal items penalise anyone testing in a second language, and that penalty measures language exposure rather than reasoning.
Sit down rested, close the other tabs, and give the set one uninterrupted run. If the conditions were bad, discount the score instead of explaining it away afterwards.
Whether taking the test again gives you a higher score
A second attempt raises scores by about four IQ points on average, with no change whatsoever in underlying ability. That estimate comes from a meta-analysis of 50 studies covering 107 samples and 134,436 participants, which found an overall practice effect of d = 0.26 on retesting (Hausknecht et al., Journal of Applied Psychology, 92(2), 2007). Which items you see the second time matters more than anything else: repeating identical items produces d ≈ .40–.42, near six IQ points, while a parallel form with fresh items produces d ≈ .22–.23, near three. By a third sitting the accumulated effect reaches about d = .51, and a later meta-analysis put the first-to-second gain at roughly a third of a standard deviation, close to five points (Scharfen, Peters & Holling, 2018).
Your first score is therefore the most informative one you will ever get from this test. Retake it if you want, subtract about five points before believing the improvement, and leave several months between attempts so item memory fades.
What an IQ score predicts about work and study
Cognitive ability predicts job performance considerably less strongly than the figure quoted on most test pages. The familiar number — a correlation near .51, from a 1998 research summary — was re-examined in 2022 and found to rest on an overcorrection for range restriction; the revised estimate is .31 (Sackett, Zhang, Berry & Lievens, Journal of Applied Psychology, 107(11), 2040–2068). A later meta-analysis restricted to studies conducted in this century, covering 153 samples and 40,740 workers, reported a mean observed validity of .16, rising to about .22 after correction.
A correlation of .31 accounts for under ten per cent of the variation in job performance. Cognitive ability remains among the better single predictors of how quickly someone learns complex work — and it shares that ground with conscientiousness, structured experience and domain knowledge, none of which appear on this page’s result screen.
For one individual the translation is blunt. Your score forecasts very little about your career, and a ±13-point band around it makes the forecast weaker still.
Whether this score can get you into Mensa
No unsupervised online test qualifies anyone for Mensa, including this one. The threshold is the 98th percentile — about 1 in 50 adults — on an approved test taken under supervision. Because scales differ, the qualifying number differs with them: 130 on the Wechsler scales and the Stanford-Binet 5, both of which use a standard deviation of 15; 132 on the older Stanford-Binet forms built on a standard deviation of 16; 148 on the Cattell III B, which uses 24. All four describe the same position in the population, and Mensa’s own published list of accepted tests is the only authority on which ones count.
A high score here is a reasonable signal that booking the supervised test is worth the fee. It is not evidence, and no chapter will treat it as such.
Why average scores rose for a century and then stopped
Population IQ climbed by roughly three points per decade through most of the twentieth century, and that climb has stalled or reversed in several wealthy countries since the 1990s. The canonical measurement pools 271 independent samples — close to four million people across 31 countries between 1909 and 2013 — and reports annual gains of 0.41 points for fluid ability, 0.30 for spatial, 0.28 for full-scale scores and 0.21 for crystallised knowledge (Pietschnig & Voracek, Perspectives on Psychological Science, 2015).
The reversal shows up inside the very item pool this kind of test draws on. Among 394,378 American adults tested between 2006 and 2018 with these items, scores fell in matrix reasoning, letter and number series, and verbal reasoning, while three-dimensional rotation scores rose over the same period. The steepest declines appeared among 18-to-22-year-olds and among people with less formal education (Dworak, Revelle & Condon, Intelligence, 98, 101734, 2023). Whatever drives the pattern, it is domain-specific rather than a general slide, and one domain moving up while three move down rules out most tidy explanations.
Two consequences for your own number. Norms carry a vintage, so a score means something only against the population it was normed on. And comparing your result with a parent’s school score from the 1980s compares two different reference populations — which is usually why the older score looks better.
When a free IQ test is the wrong tool
| If you need | Use instead |
|---|---|
| A score accepted by Mensa or another high-IQ society | A proctored test from that society’s approved list |
| An assessment connected to a diagnosis — ADHD, dyslexia, intellectual disability | A licensed psychologist administering a full battery |
| A score for a child | An age-normed instrument such as the WISC-V; adult norms do not transfer to children |
| Evidence for a gifted-programme application | Your school’s psychologist or an approved external assessor |
| To compare your score with a result from another test | The standard deviation of that scale first — 148 on Cattell III B equals 130 on Wechsler |
| A score for a job application | Whatever instrument the employer names; self-administered results are not accepted |
One case this test covers well: you want a defensible estimate of where you sit, for your own curiosity, in ten minutes, without paying or surrendering an email address. The conditions of use name the specific purposes a score from this site is not valid for, and they match the left-hand column above.
What happens to your answers
Scoring happens on your device, and no email address, account or identifying detail is required to produce a result. Nothing has to be transmitted anywhere for the calculation to run, because the scoring rules and norms ship with the page. The privacy policy inventories everything this site holds, and the cookie notice names each value stored in your browser.
Carry the same test to any assessment site you use: if a page asks for an email before showing you a score it has already computed, the score is not what is being sold.
Questions people ask before starting
What counts as a good IQ score?
Anything between 85 and 115 is statistically ordinary and covers two thirds of people; above 120 puts you in roughly the top tenth. The word “good” does more work in that question than the number does. A score describes performance on a narrow set of reasoning tasks and says nothing about judgement, originality, or how someone holds up during a hard month.
Do IQ tests measure intelligence or skill at taking tests?
Both, in a ratio that depends on the test-taker’s practice history — which is why a repeat sitting adds four to six points without any change in ability (Hausknecht et al., 2007). The stable component is nonetheless substantial: in the Lothian Birth Cohort studies, scores on the same test at age 11 and at age 79 correlated .66, rising to .73 once corrected for range restriction in the sample.
Can I take the test on a phone?
Yes, though rotation items are harder on a small screen because more of the figure has to be held in mind at once. A tablet or laptop gives that one domain a fairer reading.
Why do two tests give me different numbers?
Because each scale has its own standard deviation, its own norming sample and its own error band. A 13-point spread between two short online tests is entirely expected, and neither result is the true one — the defensible reading is wherever their confidence intervals overlap.
Sources
- Condon, D. M. & Revelle, W. (2014). The International Cognitive Ability Resource: development and initial validation of a public-domain measure. Intelligence, 43, 52–64. doi:10.1016/j.intell.2014.01.004
- Young, S. R. & Keith, T. Z. (2020). An examination of the convergent validity of the ICAR16 and WAIS-IV. Journal of Psychoeducational Assessment, 38(8), 1052–1059.
- Hausknecht, J. P., Halpert, J. A., Di Paolo, N. T. & Moriarty Gerrard, M. O. (2007). Retesting in selection: a meta-analysis of coaching and practice effects. Journal of Applied Psychology, 92(2), 373–385.
- Scharfen, J., Peters, J. M. & Holling, H. (2018). Retest effects in cognitive ability tests: a meta-analysis. Intelligence, 67, 44–66.
- Sackett, P. R., Zhang, C., Berry, C. M. & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection. Journal of Applied Psychology, 107(11), 2040–2068.
- Pietschnig, J. & Voracek, M. (2015). One century of global IQ gains. Perspectives on Psychological Science, 10(3), 282–306.
- Dworak, E. M., Revelle, W. & Condon, D. M. (2023). Looking for Flynn effects in a recent online U.S. adult sample. Intelligence, 98, 101734. doi:10.1016/j.intell.2023.101734
- Deary, I. J. et al. — Lothian Birth Cohort studies, stability of cognitive ability from age 11 to age 79.
Every factual claim on this page was checked against the source above on 25 July 2026. If something here is out of date or wrong, tell us and we will correct it with a note.
Sixteen questions, about ten minutes
Scored in your browser. No email, no card, ever.
Start the free IQ test