Nonverbal reasoning
Practice matrices in the Raven’s Progressive Matrices format
Twelve 3x3 grids with the bottom-right cell missing, eight candidates for it, and a fresh set generated every run from five published construction rules — free, no account, no clock on any item. The published items are somebody else’s property and are not reproduced here, so what comes back is a raw score out of twelve and a rule profile showing which kinds of item you solved, never an IQ number.
- 100% free
- No signup
- 12 generated items
- 8 choices, chance 12.5%
- Rules named before you start
12 matrices, generated fresh for this run from the five construction rules below, four items at each of three rule counts and eight answer choices on every one. There is no clock: the run records how long each item took but nothing is cut off.
The five rules every item here is built from
- Constant in a row
- One property holds still all the way across a row and changes when you drop to the next row. The easiest rule to see and the one that carries the opening items.
- Quantitative pairwise progression
- Something countable steps by the same amount from each cell to the next — one more element, one size up, a further 45 degrees of turn.
- Figure addition or subtraction
- The third cell is the first two combined: either everything that appears in both of them, or the first one with the second one taken out of it.
- Distribution of three values
- Three values of one property appear once each in every row and once each in every column, in an order that shifts as you go down.
- Distribution of two values
- Two values are spread across a row and the third slot gets nothing at all. The null value is what makes this the rule that separates people.
Naming the rules in advance costs the test nothing and gives you something to leave with. What makes an item hard is not knowing the list — it is holding two or three of these in mind at once while you check a candidate against all of them.
The worked examples show you the answer and name the rule that produced it. They reach no figure in the result. Answer with a click or with the number keys 1 to 8.
How to work a matrix item
Read across rows, not around the grid, and check a candidate against every rule you found.
Read the five rules on the start screen first
They are the whole grammar of the format: something holds constant along a row, something steps up by a fixed amount, two cells combine into the third, three values are dealt out one per row and column, or two values are dealt out with the third position left empty. Knowing the list does not solve an item — every item still has to be diagnosed — but it turns a grid you are staring at into a short checklist you can run.
Work row by row and name what changes
Take the top row alone and say out loud what is different between its three cells: the count, the shading, the orientation, the shape, the small mark at the top. Then check whether the middle row does the same thing. A property that behaves the same way in both rows is a rule; a property that is identical everywhere is scenery. Two worked examples are offered before the scored set and they show you the answer with its rules named, which is the fastest way to see what a rule looks like from the inside.
Test candidates against every rule you found, not the first one
The eight choices are built so that each wrong one is the correct cell with exactly one thing altered, and two of them are copies of cells already in the grid. That means a candidate can satisfy the rule you noticed and still be wrong on the rule you did not. When two rules are in play, four of the eight tiles will usually pass on one of them. Answer with a click or the number keys 1 to 8; there is no going back and no skipping.
Technical specifications
| Items per run | 12 scored matrices — four carrying one rule, four carrying two, four carrying three, presented in that order because the format is progressive by design. Two optional worked examples come first and are excluded from every figure below them |
|---|---|
| Where the items come from | Generated in your browser from a 32-bit seed at the moment you press start, using the five rules named on the page. Nothing is transcribed from any published set, and the seed travels in the copied summary so the same twelve can be rebuilt |
| Answer choices | 8 per item, so a guess is worth 0.125 and twelve guesses average 1.5 correct. Seven distractors are the correct cell with exactly one attribute changed, or a copy of the cell to the left, above or at the center of the grid — the repetition error the published format also uses |
| Attributes a rule can govern | Element count 1-5, shape from a set of eight, shading in three steps, size in five steps, orientation in 45-degree steps, an extra mark at the top of the cell, the lattice position of a single element, and the occupied-position set itself for the addition and subtraction items |
| Time limit | None. Each item's duration is recorded from the frame that painted the grid to the timestamp on your click, and the result reports the median and the total, but nothing is ever cut off |
| What voids an item | This tab going to the background or losing focus while a grid is open, an answer arriving within 150 ms of the grid painting, and any item still open when you stop the run. A voided item is named in the result and a replacement is queued at the end |
| Reference figures printed | Flynn effect, ≈ 3 IQ points — Flynn (1987), Massive IQ gains in 14 nations: What IQ tests really measure, Psychological Bulletin; and the mensa qualifying percentile, 98th percentile. Both are context for why no score appears on an IQ scale here, not something your result is scored against |
| What leaves the page | Nothing at all during a run. The copy button writes a plain-text summary and a seed token to your own clipboard, and that summary carries no per-item answers |
Frequently asked questions
Are these the actual items from the published test?
No, and they could not be. The item set is copyrighted and it is also access-controlled, which is the more interesting half: a matrix item that has circulated on the open web no longer measures reasoning in the people who have seen it, so publishing the real bank would destroy the thing it measures. What is public is the analysis of how the items are built. This page uses that analysis as a generator and says so in the results panel, where a score from generated items is stated as not comparable to a published norm.
Why is there no IQ number at the end?
Because an IQ number is a rank against a standardization sample, and there is no standardization sample behind a web page. Converting a raw score into an IQ needs a norm table collected under supervised conditions on a defined population, those tables are licensed rather than published, and matrix reasoning happens to be the ability where scores drifted upward fastest through the twentieth century — so an old table would misplace you even if you had one. The raw score and the rule profile are the two things this run can actually support.
Is this the Standard, the Advanced or the Colored version?
None of the three, and the distinction is worth knowing before you take one for a job. The family has a colored form aimed at young children and adults with significant impairment, a standard form of five sets rising in difficulty, and an advanced form built to separate people at the top end where the standard form ceilings out. The generated items here span one, two and three rules per item, which puts them across the middle of that spread rather than inside any one published form, and there are twelve of them rather than sixty.
Why eight answer choices instead of six?
Eight is what the advanced form uses, and more choices make a guess worth less. With six options an unsolved item pays 0.167 of a mark and a twelve-item run gives away two marks to luck; with eight it pays 0.125 and gives away one and a half. On a short set that difference is most of the noise. The results panel prints the exact probability of reaching your score or better by guessing every item, which is the honest way to read a low score on a twelve-item test.
Can I use a score from here to apply to Mensa or to put on a CV?
No. Admission to a high-IQ society is granted on a supervised administration of an approved test, and the criterion is a percentile rather than a number, which is why the thresholds quoted for different scales disagree with one another. Nothing taken unsupervised in a browser tab qualifies for anything, including this. What a run here is good for is finding out what the format feels like and which kind of rule you keep losing.
Why does the run not put a clock on each item?
Because a clock changes the ability being measured, and this page has a sibling that puts one on deliberately. Under time pressure a matrix test measures how fast you can cycle through candidate rules; untimed, it measures whether you can find them at all. Published administrations exist in both forms and they do not correlate perfectly. The item durations are recorded here anyway, so you can see where you spent the time without the time spending you.
The grid has changed since I last took it — is that a bug?
No, every run generates twelve new items from a new seed. That is a deliberate cost: it means you cannot learn this page, and it also means two runs are not strictly the same test, so a two-mark difference between them is at least partly which items came up. If you want to repeat a specific set, the copied summary carries the seed that produced it.
What the progressive matrix format is, and what a browser can honestly do with it
The format goes back to a preliminary paper by Penrose and Raven in 1936 and the standard series published two years later, and its design goal was to strip a reasoning test of everything except the reasoning: no words, no arithmetic, no cultural references, just a matrix of figures with one cell missing. That ambition was never fully met — schooling and familiarity with abstract diagrams both show up in scores — but the format got closer than anything before it, which is why it survived into three published forms and into most graduate selection batteries under other names. The word “progressive” is not marketing: within each published set the items are ordered so that solving the early ones teaches the grammar needed for the late ones, and an administration that shuffled them would be a different test.
What made a browser version defensible is a 1990 analysis by Carpenter, Just and Shell, who worked through the advanced form item by item and showed that the whole set is built from five rules: a property constant along a row, a quantitative progression from cell to cell, one figure added to or subtracted from another, three values distributed through a row, and two values distributed with the third position empty. Their two simulation models differed in one respect — how many rules they could hold and manage at once — and that difference reproduced the difficulty ordering of the real items. So the thing a matrix test loads on is not pattern-spotting; it is keeping several partial hypotheses alive while checking a candidate against all of them, and the distribution of two values is the rule that separates people because its third value is an absence, and an absence is the hardest thing to notice. Those five rules are what generates every item on this page, they are printed before you start, and the results panel tells you which of them you lost.
What a browser cannot do is turn any of that into a rank. The norm tables are licensed with the test kit rather than published, and even a leaked table would misplace a visitor today, because the rise in raw performance over the twentieth century was steepest on exactly this kind of abstract item — ≈ 3 IQ points on the conventional estimate, per Flynn (1987), Massive IQ gains in 14 nations: What IQ tests really measure, Psychological Bulletin. So this page reports a count and a profile, and the pages next door take the same generator somewhere else: the matrix reasoning test reports which rule and which complexity level your errors clustered in rather than the score, the abstract reasoning test puts a single graduate-assessment clock across the whole set, and the inductive reasoning test drops the grid for a row of figures and times how long the rule took to find. If you are preparing for a hiring battery, the diagram half of it is usually mechanical reasoning and the prose half is verbal reasoning, and neither is measured by this one.
A reaction time here is the interval between the frame that painted the stimulus and the timestamp the browser attached to your key, both read from the same monotonic clock. What neither can see is the display pipeline behind it, so on a 60 Hz screen roughly 16 ms of every figure below is the machine rather than you. That is the timing floor: two numbers closer together than that are the same number, and this page reports no precision it cannot support.
That floor is negligible here and is stated anyway. An item on this page takes tens of seconds rather than hundreds of milliseconds, so one frame is a part in a thousand of the number — but the same clock is used on every timed page of this site, and a page that quietly changed its story about accuracy depending on how flattering the answer was would not be worth reading on any of them.
This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish a learning disability, an intellectual disability or giftedness. Only a qualified professional, working with more than a browser, can make that judgment.
Not affiliated with, endorsed by, or connected to the owner of Raven's Progressive Matrices. Raven's Progressive Matrices is a trademark of its owner and is used here only to name the assessment this page prepares for. The questions on this page are our own: no part of the published test is reproduced, and a score here is not comparable to a score from the real instrument.
Where the twelve matrices are built and scored
Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.
The items themselves are generated in this tab from a random seed, which is why the set is different every time and why nothing had to be downloaded to produce it. Your answers are compared against the generator that made them, in the same tab, and the comparison is thrown away when you close it.