Raven's Progressive Matrices Score: What It Means and How to Read It

Raven's Progressive Matrices is the test most people have taken without being told its name. The item is always the same shape: a grid of figures with one cell left blank, and a row of candidate pieces underneath. Pick the piece that belongs. It appears in psychology studies, in army selection, in half the online tests, and in almost every discussion of what a culture-reduced test looks like. What it does not do is produce an IQ. It produces a count, and the count means something only against a norm table. This guide explains how to interpret a Raven's Progressive Matrices score: the five grades, the percentile behind them, whether the test is timed, how accurate it is and why old norms flatter modern test-takers.

What the test is, in one paragraph

The matrices were first described in 1936 in a paper by Lionel Penrose and John C. Raven, and current editions are published by Pearson. There are three families, aimed at different parts of the ability range. The Coloured Progressive Matrices is the easiest, built for young children and older adults: 36 items in three sets of twelve (A, Ab and B), untimed, 15 to 30 minutes. The Standard Progressive Matrices is the familiar one: 60 items in five sets of twelve, labelled A to E, each set opening easily and getting harder, untimed, 20 to 45 minutes. The Advanced Progressive Matrices exists to separate people at the top, where the Standard version runs out of difficulty and everyone bunches together, the same ceiling problem that limits how high any test can report. Pearson says the untimed Advanced version spreads scores within the top 20% of the population.

Scoring is a count. Each correct item earns a mark and the marks are added into a raw score, a whole number. On the Standard Progressive Matrices it is somewhere between 0 and 60, and on its own it means nothing at all.

Is Raven's Progressive Matrices timed, and how long does it take?

In its classic forms, no. Pearson's pages give these versions:

• Coloured Progressive Matrices: 36 items, for children (about 5 to 11) and older adults, untimed, 15 to 30 minutes
• Standard Progressive Matrices and the Plus version: 60 items, ages 6 to 16 and 17 and over, untimed, 20 to 45 minutes, with classic, parallel and Plus forms
• Advanced Progressive Matrices: ages 12 to 16 and 17 and over, untimed, 40 to 60 minutes; Pearson also describes a timed administration, which it says assesses intellectual efficiency
• Raven's 2: published in 2018 in the US and 2019 in the UK, on paper or digitally, with a standard score as well as a percentile rank

A digital Raven's 2 adds a wrinkle. Each examinee is given a different set of items from a bank, so Pearson's own sample report says the digital total raw scores are not directly comparable or interpretable, and the only number that can be compared is the standard score or percentile. For a digital form, a raw-score-to-IQ chart cannot exist.

A matrix item, with the rule said out loud

The published items are copyrighted and secure, so here is one built to the same pattern. Picture a three-by-three grid. The top row holds one small circle, then two small circles, then three. The middle row holds one triangle, two triangles, three triangles. The bottom row holds one square, then two squares, then a blank.

The rule, stated as a sentence: the shape is constant across a row and the count increases by one as you move right. So the missing cell is three squares. That is the whole discipline. If you can say the rule in one sentence, you have solved the item. If you are choosing the option that "looks right", you have recognised a shape, and the test is built specifically to punish that — the distractor set almost always contains a piece that completes the visual pattern while breaking the logical one.

Real items stack two or three rules at once, but the rules themselves come from a short list:

• Progression — a feature increases or decreases in equal steps along a row or column: count, size, shading, angle.
• Distribution of three — three values of a feature appear exactly once in every row and every column, in scrambled order, like a small Sudoku.
• Addition or subtraction — the third cell is the first combined with the second, or the first with the second removed from it.
• Rotation or reflection — the same figure turned by a constant angle or flipped across a constant axis at each step.
• Superimposition — two component figures overlaid, sometimes with overlapping parts cancelled rather than kept.

Read a row, name the rule, check the rule against the other row, then find the option that satisfies it. Checking against the second row is the step people skip, and it is the step that catches a rule you invented from one example. Our walkthrough of a single matrix puzzle with the working shown takes one harder item slowly and does exactly that.

From a raw score to something interpretable

A raw score of 48 out of 60 is not a result yet. It becomes one when it meets a norm table: a record of how a reference sample of people, of a stated age, in a stated country, in a stated year, actually scored. The table converts the count into a percentile — the percentage of that sample who scored at or below you. That percentile is the real output of the test.

This is the part most online versions of the test skip or fake. Raven's has no mean-100 scale built into it the way the Wechsler scales do; some publishers supply standard-score conversions alongside the percentile tables, but the percentile is the primary currency and the standard score is derived from it. Any site that hands you a raw Raven's-style score and prints "IQ 137" underneath has quietly chosen a scale, a norm group and a year on your behalf, and told you none of the three. Our note on what to check before trusting an online score lists the questions that expose it.

If you do have a percentile and want to see what it looks like as a standard score, our percentile calculator does the conversion in both directions and shows the standard deviation it is assuming, which is the number that has to be stated for the answer to mean anything.

How do you interpret a Raven's Progressive Matrices score?

Raven's own interpretation scheme grades a score by where it falls among people of the same age. In peer-reviewed papers that cite the manual, the five grades are:

• Grade I, intellectually superior: at or above the 95th percentile
• Grade II, definitely above average in intellectual capacity: above the 75th percentile (one paper says at or above)
• Grade III, intellectually average: between the 25th and 75th percentiles
• Grade IV, definitely below average in intellectual capacity: at or below the 25th percentile
• Grade V, impaired (older papers say "defective"): at or below the 5th percentile

The grades are percentile bands within an age group, not fixed raw scores, so the same raw score can be Grade I for a child and Grade III for an adult. On a scale with a mean of 100 and a standard deviation of 15, the 95th, 75th, 25th and 5th percentiles sit at about 125, 110, 90 and 75. That is our arithmetic from the normal curve, not a Pearson conversion, and the manual itself, which Pearson sells, was not something we could read: the wording above is as two peer-reviewed papers describe it.

What about a lookup for a raw score such as 51 out of 60? Pearson publishes no adult raw-score table; the percentile tables are in the manuals. The only public raw-score figures we found describe particular groups. First-year university students averaged 22.17 (standard deviation 5.60, 506 students) on Set II of the Advanced matrices, with scores from 6 to 35 (Bors and Stokes, 1998, abstract only). In a 2026 study of 1,752 Indian schoolchildren the mean Coloured raw score out of 36 rose from 17.2 at age 8 to 19.3 at age 11. Neither is a norm table, so treat any website table that turns a Raven's raw score into an IQ with suspicion.

What it measures, and what it leaves out

Raven's is the standard example of a test loaded on fluid reasoning: working out a relationship you have not been taught, on the spot, from the material in front of you. It leans very little on stored knowledge, which is what makes it useful across languages and education systems, and it is one of the most commonly used markers when researchers want a single number for general reasoning ability.

That narrowness cuts both ways. A full assessment reports several distinct index scores — verbal comprehension, working memory, processing speed and more — because those abilities come apart within a person. Raven's reports one. It cannot tell you that someone reasons abstractly well and holds information in mind poorly, and that difference is often the useful thing to know. Our explainer on full-scale IQ against index scores sets out what a single composite hides.

It is also worth retiring the idea that the test stands outside culture. Removing words removes one large source of unfairness, and that is a real advantage. It does not remove everything: comfort with abstract diagrams, exposure to pencil-and-paper puzzles, and familiarity with the multiple-choice format itself are all learned, and they are not learned evenly. "Culture-reduced" is the honest description, and it is the one the more careful literature uses. Our page on the culture-fair style of test goes through what that reduction actually buys.

How accurate is Raven's Progressive Matrices?

The best reliability tables sit in manuals we could not read, so here is what is public. In an independent study of 180 people the full Standard matrices had an internal-consistency alpha of .96, against .80 and .83 for two nine-item short forms (Bilker and colleagues, 2012). In two studies of a 12-item Advanced short form, with 608 and 479 people, test-retest reliability was .65 and .69 and the correlation with the full-scale IQ of the Wechsler Abbreviated Scale of Intelligence was .34 and .41 (2022), which describes the short form, not the full test. For children, a 2026 study of 629 Coloured-matrices takers retested after about a year found a correlation of .52 and a gain of about 3 points. Pearson reports a .53 correlation between the Watson-Glaser critical thinking test and the Advanced matrices, one figure in a range of .39 to .70 in the same report, with no sample size or version given; see our guide to the Watson-Glaser test. A moderate link with a full battery fits a test that measures one ability where the battery measures several.

A second sitting scores higher, and not because you got smarter

Matrix tests are unusually vulnerable to practice, and the reason is the same thing that makes them elegant. The item format is narrow — a handful of rule families, endlessly recombined — so somebody who has worked through fifty items has learned the search procedure itself: read a row, name the rule, check it against a second row, then find the option that satisfies it. That is a real skill, and it transfers straight to the next matrix test. The score rises while the ability the score is meant to estimate has not moved at all.

This is why supervised administrations record whether someone has sat the test before and how recently, and why retest intervals appear in test manuals. It is also the reason a matrix test you can retake at will cannot mean what a proctored one means: by the third attempt, the thing being measured is your familiarity with the format. Our piece on the practice effect on cognitive ability tests covers how large that gain usually is and how long it persists. The size is documented: a meta-analysis of 107 samples and 134,436 people found an adjusted overall retest effect of .26 across cognitive tests, about a quarter of a standard deviation, larger when identical forms are reused or practice comes with coaching, and a 1990 study found that Raven's performance could be improved significantly in a single brief training session. Pearson sells parallel forms of the Coloured and Standard matrices for retesting, and builds each digital Raven's 2 from an item bank to limit practice effects.

Have Raven's scores risen over time? Why old norms flatter new test-takers

Pearson's classic norms are old: a 2025 paper describes the age-stratified US and British norms for the Standard matrices as dating from 1993 and 1992. Scores have moved since. James Flynn's 2009 analysis found British school children gained 14 IQ points on Raven's from about 1942 to the late 2000s, 0.216 points a year, with the gains shrinking after 1979 and turning into small losses between ages 12 to 13 and 14 to 15. A 2015 meta-analysis of 271 samples from 31 countries found annual gains of 0.41 IQ points on fluid-reasoning tests, 0.30 on spatial, 0.28 on full-scale and 0.21 on crystallized tests, decreasing in more recent decades. A person scored against decades-old norms is being compared with people who had fewer of those gains, which flatters the result. Norms also fail to travel: a 2026 study of 1,752 Indian government-school children reports 1,375 of them, 78.5%, in the bottom tenth against Pearson's India norms, and builds new age norms.

Timing changes what you are measuring

Raven's can be given with a time limit or without one, and the two administrations are not the same test. Untimed, it is closer to a pure reasoning measure: whether you can find the rule at all. Timed, speed enters the score, and speed is a partly separate ability. A person who finds every rule but finds them slowly will look worse under a clock, and whether that is the right answer depends entirely on what the score is for. The important part is that the norms and the administration must match — a raw score collected under a clock, read against untimed norms, is not a comparison of anything. Our assessment states its conditions and reports against norms collected the same way, which is the minimum any score needs before it is worth quoting. One 2006 paper found that a 20-minute timed version of the Advanced matrices was an adequate predictor of the untimed score (abstract only; it prints no correlation).

The reason Raven's has survived nearly a century is that its item format is unusually honest: everything needed to solve the item is on the page, and the rule that solves it can be written down and checked by someone else. The scoring is where the honesty gets lost, because a raw count is easy to produce and a defensible norm table is expensive. When you see a Raven's-style score, ask which version, which norm group and which year. Those three answers are the score. The number on its own is just how many you got right. For where Raven's sits among the subtests of a full IQ battery, see is IQ just pattern recognition. A 2003 trial of creatine in 45 young vegetarian adults used the Advanced matrices as its intelligence outcome; see our guide to whether creatine raises IQ.

Common questions