Are IQ Tests Accurate? What the Research Says and How to Spot a Legit Test

You finished an online test, saw “127” on the screen and felt good for about four seconds. Then the doubt kicked in: are IQ tests accurate, or did a website just flatter you? The honest answer is that “IQ test” covers two very different things, from hour-long clinical batteries that are among the most precise tools in psychology to 10-question quizzes that can’t tell 105 from 125. This guide shows you how to tell them apart, with the actual numbers.

Are IQ tests accurate? Reliability vs validity

“Accurate” actually bundles two separate questions, and psychologists keep them apart. The Standards for Educational and Psychological Testing, published jointly by the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education, define them like this:

• Reliability is the consistency of scores across repeated uses of the testing procedure. If you took the test again, or a parallel version of it, would you land in the same place?
• Validity is the degree to which evidence and theory support the way a score is interpreted for a given use. The Standards call it “the most fundamental consideration” in developing and evaluating tests.

Think of a watch that always runs exactly 10 minutes fast: perfectly reliable, still wrong. A good IQ test needs both properties. It must give you about the same score twice, and that score must reflect reasoning ability rather than luck or how good you are at searching Google.

Validity also belongs to a use, not to a test in the abstract. The same score can be valid for “roughly where do I stand?” and invalid for “does my child qualify for a gifted program?” For the basics of what an IQ score is, see our guide to what IQ is and how it is calculated.

Are IQ tests reliable? Measurement error in plain numbers

No test measures anything perfectly, so psychologists estimate how far an observed score is likely to sit from your “true” score. That estimate is the standard error of measurement (SEM). In classical test theory, as David Lane’s open textbook Online Statistics Education (Rice University) explains, it is calculated as:

SEM = standard deviation × √(1 − reliability)

A 95% confidence interval then runs from 1.96 SEM below your score to 1.96 SEM above it. On the IQ scale the standard deviation is 15 points (our IQ bell curve guide explains what that means), so reliability translates directly into points.

Here is the gold standard. According to Gary Canivez’s review of the WAIS-IV in the Mental Measurements Yearbook, internal-consistency reliability for the Full Scale IQ ranged from .97 to .98 across all 13 age groups. Plug that into the formula and the SEM is about 2.1 to 2.6 points, so the 95% interval is roughly ±4 to 5 points. That is why a careful report doesn’t stop at “112”: the manual supplies 90% and 95% confidence intervals, so a psychologist would add a 95% range of roughly 107–117.

Even that is optimistic. Canivez calls these SEMs “best-case estimates” because they leave out administration mistakes, scoring mistakes and change over time. Even the best IQ test in the world gives you a range, not a pinpoint.

Does your IQ score stay the same if you retake the test?

Mostly, with two important twists.

Twist one: practice effects. Taking the same test again tends to raise your score, even when your ability hasn’t changed. In a University of Tulsa study published in The Clinical Neuropsychologist, 54 young adults took the WAIS-IV and then retook it either 3 or 6 months later. Full Scale IQ rose by about 7 points on average, and the gain was similar at both intervals. The authors concluded that the increase reflected practice, not genuine intellectual change. Canivez’s review notes the same pattern in the publisher’s own retest data, with the biggest gains on visual puzzle-type subtests.

Twist two: age and time. A 2024 meta-analysis in Psychological Bulletin by Breit, Scherrer, Tucker-Drob and Preckel pooled 205 longitudinal studies with 87,408 participants. For a 20-year-old retested after five years, the average rank-order stability was .76. Stability was low in preschool children, rose quickly through childhood and stayed consistently high from late adolescence into late adulthood. The authors also estimated that the level of stability needed for individual diagnostic decisions (.80) can only be expected above about age 7, and only over short intervals for children. That is one reason why a single early score shouldn’t define a child; our guide to the signs of a gifted child covers how US schools handle testing.

The practical lesson: practice alone raises scores, so your first attempt at a given test is the most honest read. Retaking the same test until you like the number mostly measures how well you remember the questions.

What does an accurate IQ score actually predict?

Reliability is the easy part. Validity asks whether the number means anything outside the test room, and here the evidence is solid but not magical.

The largest synthesis on school performance is a 2015 meta-analysis by Roth and colleagues in Intelligence, covering 240 independent samples and 105,185 participants. After correcting for measurement error and range restriction, the estimated correlation between standardized intelligence tests and school grades was .54. In psychology that is a strong relationship. It also leaves most of the variation in grades to other things: effort, teaching, health, interest and plain circumstance.

So are IQ tests accurate predictors? Moderately so. An accurate IQ score tells you something real about how readily you handle abstract reasoning and new problems. It says nothing about creativity, motivation or character, and it is not a verdict on your future.

Clinical IQ tests: how the gold standard works

When people ask for the most accurate IQ test, they usually mean an individually administered clinical battery. The best known adult test is the Wechsler Adult Intelligence Scale. According to Pearson’s WAIS-5 overview, the fifth edition was published in 2024, covers ages 16 to 90, takes about 45 minutes for the 7 subtests that produce the Full Scale IQ, and was normed on samples collected in 2023 and 2024 that match the US census on five key demographic variables. Pearson’s WAIS-5 fact sheet puts the normative sample at 2,020 examinees.

What makes these tests precise is not magic questions but process:

1. A trained examiner. The WAIS-5 carries Pearson’s highest purchase tier, Level C, which typically means a doctorate in psychology or a related field with clinical assessment training, or a relevant state license.
2. Standardized conditions. Same instructions, same timing, one-on-one, with the examiner watching for fatigue, anxiety or misunderstanding.
3. Representative, recent norms. Your performance is compared with people your age from a sample built to mirror the population.

That last point is easy to overlook. Raw performance on IQ tests drifted upward for decades (the Flynn effect), which makes old norms too easy. A 2014 meta-analysis of 285 studies by Trahan and colleagues estimated the drift at 2.31 points per decade overall and 2.93 points per decade for modern Wechsler and Stanford-Binet tests. On modern tests, that means a score still calculated against 20-year-old norms can flatter you by around 5 to 6 points. Any test, online or clinical, is only as good as the comparison group behind the number.

Are online IQ tests accurate? What the research shows

Better than skeptics assume, when the test is built well. Much worse when it isn’t.

The best-documented example is the International Cognitive Ability Resource (ICAR), a public-domain item set developed by researchers at Northwestern University. In their 2014 validation paper in Intelligence, Condon and Revelle reported data from 96,958 people in 199 countries who took ICAR items online, unproctored and untimed. Three findings stand out:

• Consistency was good. The 16-item form reached a reliability of .81 and the 60-item form .93.
• Scores tracked an established test. In a separate group of 137 college students, the 16-item form correlated .81 to .82 with the Shipley-2, a brief commercial intelligence test, after statistical corrections. For context, the Shipley-2 itself correlates .85 to .86 with Wechsler Full Scale IQ.
• The items stayed hard. Even with no supervision and no clock, more than half of the 60 items were answered incorrectly more often than correctly.

The real weakness of online testing is cheating. A meta-analysis by Steger, Schroeders and Gnambs in the European Journal of Psychological Assessment pooled 49 studies with 100,434 participants and found that unproctored ability tests produced scores 0.20 standard deviations higher than proctored ones, roughly 3 IQ points. The authors attribute this to cheating, and the effect was significantly smaller for questions whose answers are hard to look up online. Visual pattern and matrix puzzles have a built-in advantage here: you can’t easily type them into a search bar.

There is a catch. ICAR is a research instrument with published data; most commercial online tests publish nothing comparable. The research shows what a well-built online test can achieve, not what any particular website does. The Standards make a related point: whoever interprets a score should know if the test was unproctored or given under nonstandard conditions.

Why a 10-question IQ quiz can’t be accurate

Short quizzes fail for two mathematical reasons.

First, resolution. A 10-question test has only 11 possible raw scores (0 to 10 correct). It can report at most 11 different scores, so nearby results like 117 and 123 often collapse into one.

Second, error. Reliability depends heavily on length, and the Spearman-Brown formula estimates how it changes when a test gets shorter or longer. Starting from ICAR’s measured reliability, here is how big the margin of error gets for a person who scores 110:

A 5-question result of 110 is compatible with anything from 91 (average) to 129 (top 3%). The formula also checks out against reality: stretching ICAR from 16 to 60 items predicts a reliability of .94, and the measured value was .93.

How to spot a legit IQ test: an 8-point checklist

You don’t need a psychometrics degree to judge an online test. Run it through these eight checks; the more it passes, the more seriously you can take the result. If you want to see what good questions look like before you start, our guide to IQ test questions walks through the main types with examples.

The “realistic results” row deserves a closer look. On a normal distribution with mean 100 and SD 15, a score of 140 or higher occurs in about 1 person in 261, and half of everyone scores below 100 by definition. If every friend who tried a site got 135, the problem is the site. Our guide to what counts as a good IQ score shows what each range really means.

Red flags: certificates, Mensa claims and surprise paywalls

Three patterns deserve extra suspicion.

“Official” certificates. A PDF certificate from a website is not an official document anywhere. The clearest proof comes from Mensa itself: Mensa International’s testing FAQ states that online tests cannot be used for admission, and that even a score at or above the 98th percentile on its own online practice challenge can’t be used to qualify. Membership requires a score in the top 2% on an approved intelligence test that was properly administered and supervised.

Too-good-to-be-true scores. Flattering results are designed to be shared, not to measure. A legit IQ test will tell plenty of people they are average, because most people are.

Hidden costs. Many sites let you start for free and charge for detailed results. That model is fine when it is upfront, and not fine when the price, the renewal or the cancellation path is buried. The Federal Trade Commission’s consumer alert “Don’t let FREE cost you” gives practical advice that applies directly: read the terms before signing up (and don’t sign up if you can’t find or understand them), find out how to cancel, mark the end of any trial on your calendar and check your card statements.

Where the IQtonics test fits

If you need a diagnosis, a school placement decision, documentation for accommodations or Mensa admission, you need a clinical or supervised test, full stop. If you want a rough sense of where you stand and which reasoning skills are your strongest, an online test can be a reasonable starting point. The checklist above applies to ours too, with one honest limit: some checks, such as who the comparison group is, you can’t verify from the outside, and we don’t publish norming or reliability data.

Here is what the IQtonics IQ test is, factually (our methodology page describes how it is built and scored). It has 40 questions across five areas (patterns, spatial reasoning, numbers, logic and memory) with a 20-minute time limit. Questions get progressively harder, and harder questions carry more points. Your result is reported on the standard scale with a mean of 100 and a standard deviation of 15, with a percentile and a cognitive profile that shows your strongest area and the areas to work on. Starting the test is free; your full results unlock with a low-cost trial membership, which auto-renews unless you cancel, so please read the terms before you pay.

And to be equally plain about what it isn’t: it is an online test, not a clinical assessment, and nothing in this article is validation data for our own test. Read your score as a rough estimate. We don’t publish reliability data for our test, so we can’t state its margin of error; as the table above shows, even well-built online tests carry a margin of several points or more.