What accuracy means for an IQ test
Psychologists judge a test on three things:
- Reliability: does it give the same result twice? Full clinical batteries reach reliability coefficients of about 0.95 or higher; a test needs at least 0.85 to 0.90 to be useful for individuals.
- Validity: does it measure intelligence rather than, say, reading speed or test-taking tricks? This is shown by correlating it with established tests and with outcomes such as school results.
- Norms: who is your score compared with? An IQ is a rank, so it is only as good as the sample behind it. Norms from a few hundred self-selected visitors are not the same as a stratified national sample.
A test can be reliable and still badly normed, which gives you a consistent but wrong number.
How big is the error margin?
Every score has a standard error, and it follows from one number, the test's reliability (r): standard error = 15 × √(1 − r). The higher the reliability, the narrower the band around your score. A full WAIS, with a reliability of about 0.97, comes out at roughly ±2.5 points. For our three forms we assume a reliability for each (the same wording as on the methodology page: assumed, not measured yet), and the bands follow from the formula.
| Form | Items | Assumed reliability | ± points (about 68%) | ± points (about 95%) |
|---|---|---|---|---|
| Quick | 15 | 0.80 | ±7 | about ±13 |
| Normal | 30 | 0.88 | ±5 | about ±10 |
| Full | 45 | 0.92 | ±4 | about ±8 |
Read it like this: with the Normal form and a result of 118, the plain reading is “somewhere around 113 to 123, most likely near 118”, and about one time in three your true level sits outside that band. A 95% range is roughly twice as wide. These reliabilities are our assumption for how well the item types should hold up; they become measured figures only once enough completed runs exist, and until then we call them what they are.
Why most online IQ tests miss
- Too few items. Ten or fifteen questions cannot separate scores precisely; the margin of error becomes 10 points or more.
- No real norms. Many sites never collected norms at all and convert raw scores with a made-up table.
- Flattering scoring. A high number keeps visitors happy and encourages sharing. If a test tells nearly everyone they are above 120, the scoring is wrong, since only 9% of people are.
- No time limits. Untimed items can be solved with patience or help, which inflates results.
- No age. Without your age, a test cannot age-norm the score.
- Recycled items. Famous puzzles that circulate online are easy if you have seen them before.
What a well-built online test can do
Reasoning items transfer well to a screen. Matrix puzzles, number series and analogies do not need an examiner, and research on unproctored online testing supports this. The International Cognitive Ability Resource (ICAR), a public-domain item set developed by psychologists, produced scores online that correlated well with established ability measures in large samples.
What an online test loses is control: you might be tired, distracted, interrupted or helped. A careful test reduces this with time limits per item, a reasonable number of items (25 or more), items that do not appear elsewhere, and checks for answers that come too fast to be real. Under those conditions the result can be a fair estimate of your band, though it remains less precise than a full clinical assessment.
Retesting is the other trap. Taking the same or a similar test again raises scores by a few points on average, because you recognise the item types and pace yourself better. That gain is familiarity, not ability. If you take several online tests in a row, your first result on a well-built one is usually the most honest estimate, and a jump of 10 points on the third attempt says more about practice than about you.
How to tell a serious online test from a quiz
- It has at least 20 to 25 items and takes 15 minutes or more.
- Items are timed.
- It asks your age.
- It says how scores were normed and on whom.
- It reports a band, a percentile or a confidence interval, not just one flattering number.
- It does not claim to diagnose giftedness, ADHD, dementia or anything else.
- It does not present itself as an official WAIS, Stanford-Binet or Mensa test. Those are protected tests given under supervision; an online version can only be modelled on their format.
How our test approaches it
Our IQ test uses 25 original items, each with its own time limit, and asks your age. Scores are converted with provisional norms anchored to the general population and refitted as completed runs come in, and the full report gives the exact score with a confidence interval. The free result is a band, not a precise number, because a 25-item test supports a band more honestly than a single figure. Details, including the limits, are on our methodology page.
What a professional IQ test costs
“A real IQ test” can mean two different things: a supervised admission test for a society such as Mensa, or a full clinical assessment by a psychologist. These are the listed prices on the organisations' own pages, read on 29 October 2026.
| Route | Listed price | What you get |
|---|---|---|
| American Mensa, local group or private testing | $60 (local group), $99 (private); the page shows a promotional price until 31 October | The Mensa Admission Test, 1 to 2 hours. A qualifying score is the 98th percentile or higher; American Mensa states that it gives no IQ score or percentile report |
| Mensa UK, supervised test (in person or online) | £49 | A supervised session of 1½ hours online or 2¼ hours in person; the top two percent are invited to join |
| University of Nevada, Reno training clinic | $300 ($150 for students) | A “psychodiagnostic assessment”; the fee page does not say which tests it includes |
| UCLA Psychology Clinic | $460 to $1,725 on a sliding scale (the waitlist was closed that day); $2,814 flat rate | A comprehensive assessment with intake, testing, interpretation, report and feedback, wider than an IQ test alone |
Two things follow. A Mensa session is cheap because it answers one question, whether you reach the top two percent, and does not set out to describe your profile. A clinical assessment costs a few hundred to a few thousand dollars because a licensed or supervised examiner administers the test one to one and writes it up. An online estimate sits in a third place: it is the quickest way to see roughly where you stand, and it says so. On this site the test is free to take; the exact score and full report are one payment, no subscription.
Is an IQ of 121 to 137 good?
Yes. On the normal curve (mean 100, standard deviation 15) an IQ of 121 sits at about the 91.9th percentile, 130 at about the 97.7th and 137 at about the 99.3rd. On the classification table used on the IQ test page, 115–129 is above average and 130–144 is high, so a result between 121 and 137 means above average for the lower half and high for the upper half.
The margin matters here. With a ±5 band around a Normal-form result, a reading of 121 to 137 should be read as “above average to high”, not as a statement that you are exactly 129 or 131. The 130 line that many gifted programmes use is a cut-off on a supervised test, and a single online result close to it cannot settle which side you are on. For what individual scores mean, see IQ 120 and the percentile chart.
When you need a professional assessment
For anything with consequences, such as a learning-difficulty assessment, a gifted programme, an employment tribunal, a disability claim or concern about memory decline, only a test given by a qualified psychologist counts. A full battery like the WAIS-5 takes about an hour or more, is administered one to one and produces a profile across several areas, not just one number. An online test is a reasonable first look and a poor substitute for that.
Sources
Condon, D. M. and Revelle, W. "The International Cognitive Ability Resource: development and initial validation of a public-domain measure". Intelligence, 2014.
American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing. AERA, 2014.
Wechsler, D. WAIS-IV Technical and Interpretive Manual. Pearson, 2008.
Prices read on 29 October 2026: us.mensa.org/join/testing, mensa.org.uk/test-your-iq, unr.edu/psychological-services-center/fees, psych.ucla.edu (Psychological Assessment & Testing).
Frequently asked questions
Are IQ tests accurate?
Professionally administered IQ tests are among the most reliable measures in psychology, with reliability around 0.95. Their accuracy is still limited to a band of a few points, and they measure reasoning, not every kind of ability.
How accurate are online IQ tests?
It varies enormously. Short, untimed quizzes without norms are close to useless. A timed test with 25 or more good items and stated norms can place you in roughly the right band, with a wider margin of error than a clinical test.
How accurate are free online IQ tests?
Price is not the issue; construction is. Many free tests inflate scores to encourage sharing. Look for time limits, an age question and a reported margin of error before trusting the number.
How accurate are paid online IQ tests?
Paying does not make a test accurate. A paid report is only as good as the items and norms behind it, so check the same signs: enough timed items, stated norms and a confidence interval.
Is there an actual accurate online IQ test?
No online test matches a clinical assessment. Some, built on research item sets or with careful norms, give a fair estimate of your band. None can diagnose or replace a psychologist.
What is the most accurate online IQ test?
There is no independent ranking. The most trustworthy ones use many timed reasoning items, ask your age, explain their norms and report a range rather than an exact number.
Are IQ tests valid?
Yes, for what they claim to measure. Scores correlate with school results, years of education and job performance, and different IQ tests agree strongly with each other. They do not capture creativity, character or practical skill.
Are IQ tests reliable?
Full clinical tests are highly reliable; retaking one after a gap usually gives a score within a few points. Short online tests are less reliable, so their results should be read as a band.
How much does a real IQ test cost?
Mensa's supervised admission tests are listed at $60 or $99 (American Mensa) and £49 (Mensa UK). A full assessment at a university training clinic runs from about $300 to well over $1,000. Prices were read on the organisations' own pages on 29 October 2026; the table above gives the sources.
Is 121 to 137 IQ good?
Yes. 121 is about the 92nd percentile, 130 about the 98th and 137 about the 99th. On our classification table 115–129 is above average and 130–144 is high, so a result of 121 to 137 reads as above average to high, allowing for a margin of about ±5 points.
Is there an accurate IQ test without email?
Yes. The Fast & Accurate test needs no sign-up or email before your limited result, which shows your score range. Your exact score and full report are one payment, no subscription. Accuracy comes from the length of the test and its error margin (±7, ±5 or ±4 points by form), not from whether you give an email address.