How the Main IQ Test works

What the test measures

The Main IQ Test samples four broad abilities from the Cattell-Horn-Carroll (CHC) model, the framework most modern cognitive batteries are built on. Logical Reasoning and Pattern Recognition tap fluid reasoning (Gf) — solving novel problems without leaning on learned content — through picture matrices, figure weights, figure series and calibrated logic and number-series questions. Spatial Intelligence taps visual processing (Gv) with cube rotations and visual puzzles. Verbal Ability taps crystallised knowledge (Gc), what you have learned and can retrieve, with vocabulary and analogy items translated and reviewed per language. Memory taps working memory (Gwm) with digit and spatial spans: shown once, then typed or tapped back from memory. The test comes in three lengths — Quick (15 items, about 8 minutes), Normal (25, about 15) and Full (40, about 25) — each with the same number of items per domain, and produces both a composite score and a domain profile.

How the adaptive engine works

The test is adaptive: after each answer the engine re-estimates your ability and picks the next item from the pool whose difficulty sits closest to that estimate. Answer correctly and the next item tends to be harder; miss one and it tends to be easier. This keeps you near the edge of your ability, where each answer carries the most information about you. The picture items are generated fresh for every session from a random seed, so no two people see the same test and the answer key never leaves the server. The benefit is precision per minute: a well-targeted short test separates ability levels about as sharply as a much longer fixed one, because the items you would obviously pass, or obviously fail, are never shown.

45 seconds per question

Each question has its own 45-second clock; for memory items the clock starts only once the material has disappeared. A question you do not answer in time counts as wrong and the next one appears, so the Quick version takes under 10 minutes, Normal under 20 and Full under 30. A clock per question keeps the test genuinely timed: nobody can sit on a hard item and look the answer up, and completion times stay comparable between takers.

How scoring works

Scoring uses item response theory (IRT). Every item has a calibrated difficulty, and the chance of answering it correctly is modelled as a function of ability, with a guessing floor for multiple-choice items. From your pattern of right and wrong answers the engine estimates the single ability level that best explains them: a correct answer on a hard item moves the estimate up more than a correct answer on an easy one, which is why two people with the same number of items right can finish with different scores. The text items were calibrated on our own completed runs; the picture and memory items start on their generator's difficulty ladder and are refitted as answer data comes in. The ability estimate is then mapped onto the conventional deviation-IQ scale, which places the mean at 100 and the standard deviation at 15 — the scale on which 85 and 115 sit one standard deviation either side of the middle. Domain scores are computed the same way from the items belonging to each domain, and all three versions report on the same scale.

Where percentiles come from

Your percentile comes from our own test-taker data, with the standard assumption that scores follow a normal distribution with a standard deviation of 15. It is not drawn from a professionally stratified, nationally representative standardisation sample; we do not have one and will not claim one. Two adjustments are applied. If you tell us your age group, we correct for the well-documented decline in fluid reasoning and speed with age, so an older taker is compared with peers rather than with twenty-year-olds. And because our takers chose to take an IQ test online and report more education than the adult population, we place 100 four points below our takers' average, our estimate of that difference. 100 then approximates the average adult rather than the average visitor. Read your percentile as a reasonable but imperfect estimate of your standing in the general population.

Confidence intervals and measurement error

Every psychometric score carries measurement error. Two sittings by the same person will differ, and a short online test carries more of that error than a two-hour supervised battery does. So we report a 95% confidence interval beside your score, computed from the standard error of the ability estimate, and ask you to read the interval rather than the point. The interval is wider for the Quick version than for Full, and it widens towards the extremes, where a short test has fewer items to discriminate with. If your score sits near a band boundary, the honest reading is that you are near that boundary — not decisively on one side of it.

What this test is not

This is an online self-assessment. It is not a clinical instrument, it produces no diagnosis, and it should not be used for any decision — educational placement, hiring, medical or legal — that calls for a formal assessment. Supervised, individually administered batteries such as the WAIS and the Stanford-Binet remain the gold standard: professionally normed, delivered by a trained examiner who can see when something is off, and able to reach abilities an unsupervised online test cannot. We are not affiliated with, or endorsed by, the publishers of those instruments.

Retesting

You may retake the test, and your score will move. Familiarity with the item formats reliably adds a few points on an early retest before levelling off, and that gain reflects recognition rather than new ability. If you want a second reading worth having, leave at least a few weeks, take it rested, at a similar time of day, on the same kind of device, and somewhere quiet. The average of two honest sittings is a better estimate of you than either one on its own.

Fairness and language

The test is offered in 20 languages. Most items are pictures — matrices, figure weights, series, cube rotations, visual puzzles and spatial spans — which need no language at all and are the same whichever language you take the test in. The Verbal Ability items are the exception, and they carry a cost worth stating plainly: vocabulary and analogy items depend on the language you take them in, so someone working in a second language can score below what their reasoning warrants on that domain. Those items are translated and reviewed per language rather than machine-dumped, and the remaining text items (logic and number series) lean on structure instead of idiom. The effect is reduced, not removed, and it is confined to one domain of five.

The Fast & Accurate IQ Test (Quick, Normal, Full)

Next to the Standard test we offer the Fast & Accurate IQ Test, a culture-fair test in three lengths: Quick (15 items), Normal (30) and Full (45). Its items are pictures only — matrices, figural analogies, series, picture completion, odd-one-out, number patterns and short visual working-memory items — and every run is generated from a fresh random seed, so no two runs share an item and the answer key never leaves the server.

Items are weighted by difficulty: five difficulty levels, worth 1 to 2 points each. The weighted raw score is converted to a deviation IQ (mean 100, SD 15) with provisional norms derived from the typical difficulty of these item types; they are replaced by norms from our own completed runs once a form has 500 of them. The confidence interval follows the form's assumed reliability (0.80, 0.88 and 0.92): about ±7, ±5 and ±4 IQ points (one standard error, about a 68 % range). Because ±7 is too wide to justify a single number, the Quick form reports a range; the Full form, with nine items per domain, also reports a score per domain. Runs that are far too fast or below chance level are flagged instead of scored.

Our promise

Every number in your report is computed from your own answers: the score, the domain profile, the percentile, the confidence interval. Nothing is decorative, nothing is generated to flatter you, and no figure is invented because it would look impressive. We do not claim norms we do not have, validity we have not established, or clinical standing we do not hold. Where our data is thin or our method has limits, this page says so in plain words. If you believe a number here is wrong, tell us and we will show you how it was produced.