Search for “online IQ test accurate” and you'll get one of two answers. The marketing answer (“Yes, our test is calibrated by experts”) is mostly noise. The cynical answer (“No, only the WAIS-IV is real”) is true but useless if you just want to know what your number means. Here's the version with the math attached.
The thing being measured
Modern IQ scores aren't a raw score on a test — they're a position on a standardised distribution. The convention since Wechsler (1939) is a mean of 100 and a standard deviation of 15. So “IQ 115” means “one standard deviation above the population mean”, which by the normal CDF puts you in the 84th percentile.
A test is “accurate” to the extent that the number it reports corresponds to your actual position on that distribution — not because the questions are hard, and not because the scoring is impressive-looking.
Three things online tests usually get wrong
1. The norming sample
Clinical instruments like the WAIS-IV are normed on stratified samples of thousands of people, demographically matched to the population the test is intended for. Most online tests norm against “people who took our test” — a self-selected sample that skews younger, more educated, and more curious than the population at large. That's fine for a relative ranking among test-takers, but it inflates absolute scores when reported on the standardised scale.
2. The ceiling and floor
A 30-item test with mostly easy items has a high floor and a low ceiling. It can't tell the difference between an IQ of 130 and an IQ of 145, because someone in either range answers nearly every item correctly. Online tests reporting IQ 158+ are usually doing so on a test bank that doesn't actually contain difficult-enough items to distinguish “very high” from “exceptional”. The reported number is noise above the ceiling.
3. The test-retest variance
Most online tests don't report a confidence interval. The truth is that even a well-calibrated short test has a standard error of measurement of ±5 IQ points. Two attempts on the same instrument can legitimately differ by 10 points without any underlying cognitive change — just from item sampling and the day's fatigue. If a site reports your IQ to the nearest integer with no confidence band, treat it as an estimate with that uncertainty built in.
What a careful online test can actually tell you
Despite all that, a thoughtfully designed online assessment is useful for:
- Your standardised position roughly within ±5 IQ points — enough to know if you're comfortably average or clearly above the population mean.
- Per-domain shape — the relative balance of pattern, numerical, verbal, and spatial reasoning. This is often more actionable than the overall number.
- Within-instrument trends — how your score on the same instrument changes over time, especially after targeted training.
What it can't
- Diagnose anything. Diagnostic assessment requires individual administration by a licensed psychologist with a validated instrument.
- Distinguish reliably above ~140 IQ. The math gets noisy at the tails.
- Compare you to a friend who took a different test. Different instruments, different norming samples — the numbers aren't directly comparable.
What we do
LabTest IQ publishes its math. The piecewise-linear raw-to-IQ curve, the Abramowitz & Stegun erf approximation for the normal CDF, the cohort z-scoring for per-domain norms — all of it is on /methodology. Items are hand-authored, difficulty is hand-calibrated, and the bank is small (30 items) by design: every item gets reviewed for clarity and single-answer correctness, no machine-generated fillers.
Take the result as what it is: a careful screening estimate, on the same canonical scale as the clinical literature. Use it to decide what to train, not to settle an argument.
If you want a real number
See a licensed clinical psychologist. The WAIS-IV (or equivalent regional instrument) administered individually is the only way to get a defensible clinical number. It costs more, takes longer, and requires booking. For most people, a careful online screening is the right level of precision for the question they're actually trying to answer.