About the test
Matrix 40 is an IQ test of 40 matrix-style puzzles with a 40-minute time limit.
The puzzles start very easy and climb steadily to very hard.
Your score and your IQ estimate appear the moment you finish the test, along with how you did on every puzzle.
The test is free, with no sign-ups or ads.
Who makes it
I'm Dagh Nielsen and I make Matrix 40 at Gamestairs, a small company in Tartu, Estonia.
In 2015 I made videos explaining the puzzles of an IQ test. The videos have accumulated 5 million views over the years. I still love these matrix-style puzzles, so I decided to make a new test.
The puzzles were created one at a time. AI did most of the designing, with me as the editor. I worked through every puzzle, discussed and revised as needed, and narrowed about 115 candidates down to the final 40.
I'm also making videos on YouTube to explain the solutions. These are my explanations, not the last word. If you see a puzzle differently, or think I've got one wrong, say so in the comments.
How your IQ is estimated
Your raw score is the number of puzzles you solve in 40 minutes. This score places you in a percentile among our players: better than 84% means that 84% of our players scored lower than you.
Every such percentile can be translated directly to a place on the famous bell curve. The bell curve has two parameters:
- The middle is put at 100, meaning that 50% of a population are assigned an IQ above 100.
- The width of the curve can be scaled as you like. Most modern tests use a standard deviation (SD) of 15, some use 16, and some, especially online, use 24, which makes the same result look higher. We use 15.
In table form, it looks like this:
| Better than, of everyone | On the curve | SD 15 | SD 16 | SD 24 |
|---|---|---|---|---|
| 50% | middle | 100 | 100 | 100 |
| 84% | +1 SD | 115 | 116 | 124 |
| 98% | +2 SD | 130 | 132 | 148 |
| 99% | +2.33 SD | 135 | 137 | 156 |
However, there's an important catch. People who choose to take an online IQ test would probably get higher raw scores, on average, than the general population. The middle player on our test may well be better than 75% of the general population. That means we can't translate our players' percentiles directly into general population IQs.
So what to do? A professional test typically settles this with a norm study, but we haven't done one. Instead we make an assumption: our players' IQs follow a bell curve with mean 110 and SD 13.
An 84% player on our test (1 SD above the middle) thus gets an IQ of 110 + 13 = 123. That's better than about 94% of the general population. Without this assumption, the same player would only get an IQ of 115.
| Better than, of our players | IQ we give | Without the assumption | Lift |
|---|---|---|---|
| 2% | 84 | 70 | +14 |
| 16% | 97 | 85 | +12 |
| 50% | 110 | 100 | +10 |
| 84% | 123 | 115 | +8 |
| 98% | 136 | 130 | +6 |
| 99.9% | 149 | 145 | +4 |
Why 110 and 13
No study we found places people who take online IQ tests on the general population's scale, so both numbers are estimates.
The mean. In a 2013 Pew survey, 69% of visitors to an online news quiz were college graduates, against 29% of the public. Graduates score about 14 points above others (worked out from Strenze 2007), so a group like that averages about 106. People who seek out a 40-minute IQ test are likely more selected than quiz visitors, so we assume 110.
The SD. A group that selects itself usually varies less than the general population. 13 is our judgement of how much less.
We'll check both of these assumptions by comparing our results with the scores players report from other tests, and we'll correct them if needed.
Some notes
- Tests with under 10 minutes spent on the puzzles don't count.
- Only first attempts count towards the percentile, as practice raises scores.
- No test measures exactly. Read your IQ as give or take about 5 points, one standard error of measurement for a test of this kind.
- Few players and few puzzles separate the highest scores, so read a number above about 145 as "very high", not as an exact value.
- We don't adjust for age. Raw scores fall through adult life, so this number understates older players against people their own age.
- Matrix 40 isn't a clinical assessment.
IQ tests in general
Nearly all the research below comes from long tests given under supervision. No study has yet checked Matrix 40 against real life, so none of this is a claim about our test.
Why IQ tests exist
The first was made in 1905 by Alfred Binet and Théodore Simon, for a French government commission looking for a way to find children who needed special teaching (Minton). Psychologists still use such tests for work of that kind, such as assessing intellectual disability or the effects of a brain injury. Employers and armed forces use them to select people. And plenty of people take them out of curiosity, or because the puzzles are fun.
What the research finds
Mental abilities overlap. People who do well on one kind of mental test tend to do well on others, whether it uses words, numbers, shapes or memory (Warne & Burningham 2019). A statistical summary of this overlap is called g, and what lies behind it is still debated. Matrix tests track g well but not purely, so part of any matrix score belongs to matrix puzzles alone (Gignac 2015).
Scores predict real outcomes, on average. Higher scores go with more education, better jobs, higher pay and even a longer life (Calvin 2017). A correlation measures how closely two things go together, from 0 (not at all) to 1 (perfectly). To see what these correlations mean, pick two people at random. How often does the one with the higher score also come out ahead?
| Also ahead on | How often | Correlation |
|---|---|---|
| Education | 69% | .56 |
| Job status | 64% | .43 |
| Job performance | 57–60% | .22–.31 |
| Pay | 56% | .20 |
A coin flip would be 50%. The correlations come from Strenze 2007 and Sackett 2022, 2023, as corrected there. The percentages are our own rough conversion.
These are strong results by the standards of psychology (Funder & Ozer 2019), but they're still only averages.
Scores are fairly stable, but not fixed. From the late teens on, people's scores relative to each other change fairly little over the years (Breit 2024). Still, each extra year of school adds about 1 to 5 points (Ritchie & Tucker-Drob 2018), and whole generations have moved: Dutch men born in 1964 scored about 20 points above men born in 1934, on a matrix test (Flynn 1987). So a matrix test isn't culture-free.
What a good test claims, and what it doesn't
A good IQ test claims something modest: it estimates how you compare with others at certain kinds of reasoning, and that estimate relates, on average, to some things in life. It doesn't claim:
- To measure your worth. A score says nothing about your kindness or your character, or about what you've done with your life.
- To capture everything that counts. Personality traits predict job status, divorce and lifespan about as well as test scores do (Roberts 2007).
- To be exact. Besides the give or take above, scores move with practice. Taking a test again adds about 4 points on average, and more with coaching (Hausknecht 2007). Videos that explain matrix puzzles count as coaching, ours included.
- To tell your future. Look at the table again: the person with the higher score earns more only a little over half the time.