What is fluid intelligence, and how is IQ measured?

Most people who search for a free IQ test want two things: a number, and some idea of what that number means. This guide covers the second part. It explains what kind of ability a puzzle-based IQ test actually measures and how a raw score becomes an IQ. It also explains why any single result — especially from a short online test — is best read as a range rather than a verdict.

Fluid and crystallized intelligence

In the 1960s the psychologist Raymond Cattell argued that “general intelligence” is better described as two related but distinct abilities (Cattell, 1963):

• Fluid intelligence is the ability to reason through new problems: spotting patterns, working out rules and drawing conclusions when prior knowledge gives you little help.
• Crystallized intelligence is the knowledge and skill you have built up through education and experience, such as vocabulary, general knowledge and familiar ways of solving problems.

The two usually go together, but they are not the same thing. A well-read person may have a large vocabulary (crystallized) and still find an unfamiliar logic puzzle hard (fluid), and the reverse also happens.

Cattell’s split later became part of a broader framework known as the Cattell–Horn–Carroll (CHC) model. CHC pulls together a century of research on cognitive abilities. It places a general factor at the top, a set of broad abilities in the middle — fluid reasoning, comprehension-knowledge, visual processing, processing speed and others — and many narrow, specific skills underneath (McGrew, 2009). It is widely accepted as a framework for designing and interpreting intelligence tests.

The practical point: “IQ” is not one fixed substance. Different tests sample different abilities, and a test built from visual puzzles leans heavily on fluid reasoning.

What matrix reasoning tasks are

A matrix reasoning item shows a grid of shapes — often 3×3 — with one cell missing. The shapes change across rows and down columns according to hidden rules: a shape might rotate, grow, gain an extra copy, or swap fill patterns. Your job is to work out the rules and choose the piece that completes the grid.

The best-known example is Raven’s Progressive Matrices, developed by John C. Raven in the 1930s and since used in many countries (Raven, 2000). The items are “progressive” because they start simple and get harder, and they use abstract shapes instead of words or numbers.

Matrix tasks are popular for three reasons:

• They target fluid reasoning. Because the shapes are unfamiliar, there is little stored knowledge to fall back on; you have to find the rule on the spot.
• They need little language. That makes them easier to use across languages, although no test is completely free of cultural influence (Raven, 2000).
• They are efficient. Each item is quick to present and has one clear correct answer.

Commercial tests such as Raven’s are copyrighted, so researchers created an open alternative. The International Cognitive Ability Resource (ICAR) is a public-domain set of item types: matrix reasoning, letter and number series, verbal reasoning and three-dimensional rotation. It was tested with tens of thousands of online participants. Its developers found that making the items freely available did not appear to undermine their validity (Condon & Revelle, 2014). ICAR is intended for research use, and its items are not reproduced here or on our site.

How IQ scores are scaled

The number of puzzles you solve is a raw score. On its own it means little: 12 correct is impressive on a hard test and ordinary on an easy one. To make scores comparable, tests convert raw scores using norms — the distribution of scores in a reference group of people who took the same test.

Modern IQ tests set the average of that reference group at 100 and the standard deviation at 15 (Neisser et al., 1996). Because scores in a large population roughly follow a bell curve, each IQ value corresponds to a percentile — the share of people expected to score lower:

• IQ 100 — 50th percentile (exactly average)
• IQ 115 — about the 84th percentile (one standard deviation above)
• IQ 130 — about the 98th percentile (two standard deviations above)
• IQ 85 — about the 16th percentile (one standard deviation below)

Roughly two-thirds of people score between 85 and 115. A difference of a few points is therefore small; a difference of 15 is substantial.

Measurement error and confidence intervals

No test measures anyone perfectly. Your score on a given day is affected by sleep, attention, the particular items you saw and a fair amount of luck when guessing. Psychologists describe this with a reliability figure: the closer it is to 1, the less a person’s score would change if they took a parallel version of the test.

From the reliability you can estimate a confidence interval — a range likely to contain the score you would get on average if you could take the test many times. A careful report gives you this range, not just a single number. If your result says “108 (90% range 97–119)”, the honest summary is “somewhere around average to above average”, not “exactly 108”.

What online and short tests can and cannot tell you

Professional IQ assessments are usually given in person by a trained examiner, cover several broad abilities rather than one, and are normed on large reference groups. A free online test usually differs in all of those respects.

A short online matrix test can:

• give you a rough sense of how you handle abstract pattern-finding under time pressure;
• show which kinds of rules you find easy and which you find hard;
• be an enjoyable puzzle session and a starting point for reading about intelligence research.

It cannot:

• give you a full-scale IQ, because it samples only one ability;
• separate fine differences, especially at the very top or bottom of the range, because there are few items;
• control for the testing conditions — distractions, a small phone screen, or tiredness all lower scores;
• serve as a clinical, educational or employment assessment.

It is also worth keeping in mind what IQ predicts in general. Scores correlate with school performance at about .50 (Neisser et al., 1996) — a meaningful link, but one that leaves most of the differences between students explained by other things, such as motivation, teaching and circumstances.

Genes, environment and the Flynn effect

Twin and family studies find that differences in intelligence between people are partly heritable, and that the heritable share rises with age — from about 20% in infancy to perhaps 80% in later adulthood (Plomin & Deary, 2015). Heritability describes variation across a population; it does not say that any one person’s score is fixed or “mostly genetic”.

The environment clearly matters too. A meta-analysis of studies designed to separate cause from effect estimated that each additional year of education raises intelligence test scores by about 1 to 5 points (Ritchie & Tucker-Drob, 2018).

The clearest sign of environmental influence is the Flynn effect: average scores on intelligence tests rose throughout the 20th century. A meta-analysis covering about four million people from 31 countries, between 1909 and 2013, found gains of roughly three IQ points per decade on full-scale tests, with larger gains on fluid reasoning tests than on crystallized ones, and a slowdown in recent decades (Pietschnig & Voracek, 2015). Raven’s own norms show the same pattern: fairly similar across cultures at any one time, but shifting substantially over the decades (Raven, 2000). This is one reason norms must be updated — a test normed long ago will make today’s test-takers look better than they are.

Common myths

“An IQ score is a fixed, permanent number.” Scores have measurement error, change with education, and population averages have shifted over time (Ritchie & Tucker-Drob, 2018; Pietschnig & Voracek, 2015).

“Brain-training games will raise my IQ.” Working memory training improves performance on the trained tasks, but a meta-analysis found no convincing evidence that the gains carry over to broader reasoning abilities (Melby-Lervåg & Hulme, 2013).

“IQ measures everything that matters about a mind.” IQ tests sample certain reasoning and knowledge skills well. They do not measure creativity, practical wisdom, personality or many other qualities that shape how a life goes (Neisser et al., 1996).

“A free online test gives you your real IQ.” It gives you an estimate of one ability, on one occasion, with a wide margin of error.

How our IQ test works

The Mind Examination IQ Test (Quick) is a short fluid-reasoning test built entirely from matrix puzzles. Here is exactly what it does:

• Items. 16 original 3×3 matrix puzzles, plus one practice puzzle. They are generated from rules we wrote ourselves; nothing is copied from Raven’s, ICAR or any other published test.
• Difficulty. Four levels of four puzzles each. Level 1 has one changing feature, such as the number of shapes. Higher levels combine two or three rules at once, or use line patterns that are added together, cancel out or overlap.
• Time. 10 minutes in total. Wrong answers cost nothing, so it pays to guess rather than leave an item blank.
• Raw score. The number of correct answers, from 0 to 16.
• Scaling. The raw score is converted to an estimated IQ (mean 100, SD 15), shown with a 90% range and a percentile.

The norms are provisional. Until enough people have taken the test, we use an estimated average of 9 correct out of 16 with a standard deviation of 3.2, based on what is typical for a 16-item matrix test of balanced difficulty. We also assume a reliability of 0.80, which gives a 90% range of about 11 points either side of your estimate. Once enough anonymous results have been collected, we will recalculate the norms from real site data and note the change on our Methodology page.

Two consequences follow from the design. First, each extra correct answer moves the estimate by roughly 4–5 points, so a single lucky guess or slip matters. Second, with only 16 items, the scale cannot tell apart very high or very low scorers precisely: a perfect score corresponds to an estimate in the low 130s, and estimates are capped at 55 and 145. Treat your result as a rough, enjoyable snapshot of your pattern-reasoning on the day — not a measurement of your worth or potential.

Frequently asked questions

What does an IQ score of 100 mean?

It means you scored exactly at the average of the reference group the test was normed on — the 50th percentile. About half of people score higher and half lower. On any test, the range around your score matters as much as the number itself.

Is 120 a good IQ score?

An IQ of 120 is about 1.33 standard deviations above average, which corresponds to roughly the 91st percentile. On a short online test, though, the confidence interval is wide, so a 120 on our test is better read as “somewhere between about 109 and 131”.

Why did I get different scores on different IQ tests?

Tests measure different mixes of abilities, use different norm groups, and have their own measurement error. Old norms also inflate scores because of the Flynn effect (Pietschnig & Voracek, 2015). Differences of 10 points or more between two unrelated online tests are common and not surprising.

Can I improve my fluid intelligence?

Practicing a specific kind of puzzle will make you better at that puzzle, but training games have not been shown to raise broad reasoning ability (Melby-Lervåg & Hulme, 2013). The most consistent evidence for raising measured intelligence comes from education (Ritchie & Tucker-Drob, 2018).