Are IQ tests accurate?

Short answer: A full IQ test given by a psychologist gives very consistent scores. On the WAIS-5, the Full Scale IQ has an average reliability of 0.97 (Winter, Trudel and Kaufman, 2024, citing the WAIS-5 manual). Calculated from that figure, a score's 95% range is about 5 points either side. Online tests vary. One short research test, ICAR16, tracked WAIS-IV scores closely (.81) in a study of 97 students (Young and Keith, 2020). Most free online quizzes publish no evidence of accuracy at all. Any single score is an estimate with a range around it.

How accurate each kind of IQ test is

Here is what the published evidence says for each kind of test, from the most to the least studied:

• A full clinical test, such as the WAIS-5, given in person by a psychologist. Full Scale IQ reliability of 0.97, and 0.93 when 201 people took it twice about a month apart (Winter, Trudel and Kaufman, 2024, citing the WAIS-5 manual). Best for any decision about school, work or health.
• The same clinical test given by a psychologist over video. In a 2025 study of 28 adults who took the WAIS-IV both ways, the differences were smaller than measurement error alone would produce (Bartholomaeus and colleagues, The Clinical Neuropsychologist, 2025). The sample is small, and the study did not formally show the two scores were equivalent.
• A short test built and checked by researchers, such as ICAR16. In 97 students, the 16-item ICAR16 correlated .81 with WAIS-IV Full Scale IQ (Young and Keith, 2020). The authors call for larger studies, and the test is meant for research use only (ICAR project).
• A short matrix test with published data. Two 9-item versions of Raven's Standard Progressive Matrices had reliabilities of .80 and .83, against .96 for the full 60 items (Bilker and colleagues, Assessment, 2012). Shorter tests give wider ranges.
• A free online quiz with no published data. We found no study that measures how accurate these are. We searched Europe PMC and the open web on 7 October 2026. Without published norms or reliability, there is no way to check the number it gives you.

What "accurate" means for an IQ test

People use "accurate" for two different things, and psychologists keep them apart.

Reliability is whether a test gives you about the same score each time. A 2020 guide in Simulation in Healthcare puts it simply: reliability is "the degree to which a measure can produce stable and consistent results".

Validity is whether the score means what people say it means. The testing profession's own standards (AERA, APA and NCME, 2014) define it as "the degree to which evidence and theory support the interpretations of test scores for proposed uses of tests." Validity belongs to a particular use of a score. A test can be valid for comparing reasoning ability and not valid for choosing a career.

So "is this IQ test accurate?" asks two things at once: whether you would get about the same score again, and whether the score means what people say it means.

How accurate clinical IQ tests are

Clinical tests have the most published data, and their consistency is high.

• WAIS-IV (2008): Full Scale IQ reliability of 0.98, and 0.96 when 298 people took the test twice an average of 22 days apart (Collinson and colleagues, Journal of Intelligence, 2016, summarizing the WAIS-IV manual).
• WAIS-5 (2024): Full Scale IQ reliability averaging 0.97 across ages, and a retest figure of 0.93 for 201 people aged 16 to 90 (Winter, Trudel and Kaufman, 2024).
• Part scores: the separate index scores ranged from 0.90 to 0.98 on the WAIS-IV and 0.90 to 0.97 on the WAIS-5, according to a 2025 study in the Journal of Intelligence. A part score is usually a little less precise than the full score.

A reliability number is easier to read once you turn it into points. The standard formula multiplies the scale's standard deviation (15) by the square root of one minus the reliability, which gives the standard error of measurement. The traditional way to get a 95% range is the score plus or minus 1.96 of those errors (Schmukle, Assessment, 2025). Here is what that gives, calculated from the figures above:

• At a reliability of .97 (WAIS-5), the 95% range is about 5 points either side. A measured 128 most likely sits between about 123 and 133.
• At .90, it is about 9 points either side.
• At .80, it is about 13 points either side, and at .83 about 12.

Psychologists report scores this way too. Pearson's WAIS-5 report software asks the clinician to "Select the desired confidence interval (i.e., 90% or 95%)."

Scores also hold their order over long periods. In a 2024 meta-analysis of 205 studies and 87,408 people, the stability of cognitive ability scores over 5 years was .76 for a 20-year-old (Breit and colleagues, Psychological Bulletin, 2024). In one group of 106 people, the same test taken at age 11 and again at 90 correlated .54, or .67 after a statistical correction (Deary, Pattie and Starr, Psychological Science, 2013). These are long-term figures. Retest figures over a few weeks are the ones in the list above.

Are online IQ tests accurate?

Some can be, many cannot be checked, and "online" alone does not decide it. Two separate questions get mixed up here: whether anyone supervises you, and whether the test itself has been checked.

Being unsupervised raises scores a little. A meta-analysis of 49 studies from 2001 to 2017 found that people scored slightly higher without a supervisor, a standardized difference of d = 0.20 (Steger, Schroeders and Gnambs, reported by Scherrer and colleagues, Journal of Intelligence, 2025). On the IQ scale that is about 3 points, calculated. The gap was largest on tasks you could look up online. In Scherrer's own 2025 study of 226 people, supervision made no significant overall difference, and the authors concluded it was "less critical for complex cognitive tasks" such as reasoning puzzles. A test publisher, Hogrefe, funded that study.

Clinical tests given online by a psychologist look close to in-person testing in two studies. In addition to the WAIS-IV video study above, a 2020 study of the children's WISC-V found no significant differences in Full Scale IQ between online and in-person administration (Wright, Psychological Assessment, 2020). One WISC-V subtest scored lower online.

Research tests can work on a phone. In a 2025 study of 100 adults, 63 retook an ICAR matrix test remotely on their own phones 11 to 17 days after a first session in the lab, and the retest correlation was 0.82 (Young and colleagues, Journal of Intelligence, 2025). The same test correlated 0.40 with Raven's Progressive Matrices, which the authors describe as a moderate to strong correlation.

Most free quizzes cannot be checked. You may read that free online tests inflate scores by a set number of points. We found no published study that measured it, so treat any such figure with care. A quiz with no published data has no evidence for or against its accuracy.

American Mensa is strict about this. It says: "We do not accept unsupervised testing as proof of eligibility, specifically unsupervised testing administered electronically, or via internet-based tests."

Why a shorter test gives a wider range

Each puzzle in a test is a small sample of how you reason, and more puzzles give a steadier estimate. Ellis and Sijtsma explain that reliability rises toward its maximum as a test gets longer, at the rate the Spearman-Brown formula predicts (Psychometrika, 2024).

The Raven's short forms show the cost. The full 60-item Standard Progressive Matrices had a reliability of .96, and the two 9-item versions had .80 and .83 (Bilker and colleagues, 2012). The short forms still ranked people much like the full test, correlating about .90 with it in new data. By the formula above, reliabilities of .80 and .83 mean a 95% range of about 13 and 12 points either side, calculated, against about 5 for a full clinical test. A short test can be useful, as long as it shows you a wider range.

What can make your score less accurate

Even a well-built test can give one person a misleading number. These are the effects with the clearest evidence:

1. You have taken it before. On the WAIS-IV, retaking the test raised Full Scale IQ by about 4.5 to 5 points for young and middle-aged adults, and on the WAIS-5 by 3.7 points for ages 16 to 69 (Winter, Trudel and Kaufman, 2024). A 2007 meta-analysis of 134,436 people found practice effects were larger with coaching and with the identical test (Hausknecht and colleagues, Journal of Applied Psychology). Gains tend to level off by about the third sitting (Scharfen, Peters and Holling, 2018, as summarized in 2019).
2. You were not trying hard. In a 2011 meta-analysis of experiments with 2,008 people in low-stakes research settings, material incentives raised IQ scores by an average of 0.64 standard deviations (about 10 points, calculated), with larger gains for people who started with lower scores (Duckworth and colleagues, PNAS, 2011). Take a test when you can give it your full effort.
3. The norms are old. The APA task force report (1996) put the average gain in scores at about three IQ points per decade, and a 2014 meta-analysis put it at 2.93 points per decade for modern Stanford-Binet and Wechsler tests (Trahan and colleagues, Psychological Bulletin). A test compared with an old norm group can read high. Two studies in the WAIS-5 manual found a gain of about 1.2 points per decade where 3 were expected, in small samples (Winter, Trudel and Kaufman, 2024).
4. Someone helped, or you looked things up. Unsupervised scores run highest on tasks you can search (Scherrer and colleagues, 2025). Take the test alone and without looking anything up.

What IQ scores predict, and what they miss

A test can be consistent and still only tell you part of the story. Scores do predict some real outcomes:

• School grades: the APA task force report (1996) put the correlation between IQ and grades at about .50, which accounts for "only about 25% of the overall variance". A 2015 meta-analysis of 240 samples and 105,185 people found .54 after statistical corrections (Roth and colleagues, Intelligence, 2015). The APA report also names "persistence, interest in school, and willingness to study" as part of learning.
• Job performance: older estimates put it at .51 to .54 (Hunter, 1983, in the APA report; Schmidt and Hunter, 1998). In 2022, Sackett and colleagues revised it down to .31 (Journal of Applied Psychology), which accounts for about 10% of the differences in job performance, calculated. They concluded that earlier corrections for range restriction had overcorrected.

There is also a lot a test does not try to measure. The APA report says: "Obvious examples include creativity, wisdom, practical sense and social sensitivity; there are surely others." Some researchers question the idea of one number altogether: Hampshire and colleagues (Neuron, 2012) argued that intelligence comes from several distinct brain systems, each with its own capacity. Treat that as one view in an open debate.

If you want to know what your number means on the scale, see what the average IQ is and how scores spread. For the kind of reasoning matrix puzzles test, see what fluid intelligence is.

How to tell whether an online IQ test is accurate

Use these checks before you trust a number from an online test:

1. It names its method. It says what kind of puzzles it uses and which scale it reports on, such as the standard IQ scale, which has an average of 100 and a standard deviation of 15.
2. It shows a range around the number. Every score carries error, so an honest test shows the range your score most likely sits in.
3. It publishes reliability or validation data, or says plainly that it has none yet. If a test says it is accurate, look for the numbers that back it up.
4. It says who it compares you with. A score only means something against a norm group, so look for who that group is and when it was measured.
5. It does not call itself official, certified or clinical. No online quiz replaces an assessment by a psychologist.
6. It does not promise a high score. Look for a test that reports the score your answers earn, whatever it is.
7. You took it alone, trying your best and without help. Your own conditions matter as much as the test.

When you need a clinical assessment instead

For any decision that matters, such as school placement, a question about learning or attention, or a health concern, see a qualified professional. The American Psychological Association says: "Licensed clinical psychologists are expertly trained to administer assessments and tests and interpret the results." A full assessment can combine several tests with interviews, records and observation, chosen for the person (APA, 2022). Tests like the WAIS-5 are sold only to qualified professionals under Pearson's level C rules.

An online test is useful for a different job: a quick, low-stakes estimate of how you do on one kind of reasoning. If that is what you want, WhatIQ's test has 24 timed picture puzzles and takes about 15 minutes (whatiq.com, read 7 October 2026). It gives an estimate rather than a clinical assessment and shows your score with the range it most likely sits in. WhatIQ's terms, last updated 6 October 2026, say it is still measuring how hard each puzzle is and will adjust the scoring as more people take the test.

Frequently asked questions

Are online IQ tests accurate?

It depends on the test. Clinical tests given online by a psychologist scored close to in-person testing in a small 2025 study (Bartholomaeus and colleagues, 2025) and in Wright's 2020 study of the children's WISC-V, and one research test, ICAR16, tracked WAIS-IV scores closely in a study of 97 students (Young and Keith, 2020). Most free quizzes publish no data, so their accuracy cannot be checked.

Are free online IQ tests accurate?

A free test can be accurate, but most free quizzes publish no reliability or norm data, so there is no evidence either way. Check whether the test names its method, shows a range and explains who it compares you with. The ICAR tests are public-domain tests with published studies behind them, though they are meant for research use.

Why do online IQ tests give such high scores?

A test taken without supervision scores about 3 points higher on average, calculated from d = 0.20 in Steger and colleagues' meta-analysis (reported by Scherrer and colleagues, 2025). Practice from earlier attempts and outdated norms can push a score up too. A test never checked against a norm group can give a number that is too high or too low, and we found no study that measured how much free quizzes inflate scores.

How accurate is the WAIS?

Very consistent. On the WAIS-5, the Full Scale IQ has an average reliability of 0.97 and a retest figure of 0.93 (Winter, Trudel and Kaufman, 2024). Calculated from the 0.97 figure, a WAIS-5 score's 95% range is about 5 points either side, so psychologists report it as a range.

Can your IQ score change if you retake the test?

Yes, usually by a few points. Retaking the WAIS-5 raised Full Scale IQ by 3.7 points on average for ages 16 to 69 (Winter, Trudel and Kaufman, 2024). Across many ability tests, gains tend to level off by about the third attempt (Scharfen, Peters and Holling, 2018, as summarized in 2019). How hard you try and how old the test's norms are can move a score too.

Is the online Mensa test accurate?

Not for Mensa's purposes. American Mensa does not accept unsupervised or internet-based tests as proof of eligibility, so no score from a test you take online on your own counts toward joining. Mensa's rule is about proof of eligibility, not a measure of any one test's accuracy.