MindProofMindProof · Puzzle guides
What is a good IQ score?
What the numbers on an IQ scale actually mean, the percentile each one corresponds to, and why a single score is less precise than it looks.
Most IQ scales are built so that the average is 100 and one standard deviation is 15 points. That is a design decision, not a discovery: the scale was defined that way so scores could be compared against a reference group. A score is therefore not a measurement of a quantity in your head. It is a statement about where your performance sat relative to the people the test was normed on.
That framing matters for reading the number honestly, so it is worth starting there rather than with a table.
What each score corresponds to
Because the scale is defined around a normal distribution, the percentile for any score is arithmetic rather than opinion. These figures follow directly from a mean of 100 and a standard deviation of 15:
| Score | Standard deviations | Percentile | Roughly |
|---|---|---|---|
| 70 | −2.0 | 2nd | about 2 people in 100 score at or below |
| 85 | −1.0 | 16th | about 16 in 100 |
| 100 | 0 | 50th | the midpoint by definition |
| 115 | +1.0 | 84th | about 84 in 100 score at or below |
| 130 | +2.0 | 98th | about 2 in 100 score above |
| 145 | +3.0 | 99.9th | about 1 in 750 score above |
Two consequences follow. About 68% of people score between 85 and 115, and about 95% score between 70 and 130. The range that feels unremarkable is where most of the population sits, and small differences near the middle of the scale separate far more people than the same difference out at the tail.
So what counts as good?
Above 100 is above the reference average, and that is the entire defensible answer. Any stronger claim smuggles in an assumption about what the score is for.
A score is only meaningful alongside three things: which test produced it, which reference group it was compared against, and how much uncertainty the test reports. A number quoted without those is not interpretable, however precise it looks.
Not every scale uses 15
Some instruments use a standard deviation of 16, and a few historical scales used 24. The same raw performance produces a different-looking number on each. This is one reason scores from different tests should not be compared directly, and why a score quoted with no test attached tells you very little.
Why one number is less precise than it looks
Every test has a standard error of measurement: the spread you would expect if the same person took equivalent versions repeatedly without learning anything in between. That is why well-constructed reports give a confidence band rather than a bare number. A result reported as a single integer is presenting more precision than the underlying measurement supports.
Several ordinary things move the number without anything changing about you:
- Familiarity. Having seen the item format before helps, which is why a second attempt at the same test is a weaker measurement than the first.
- Conditions. Tiredness, interruptions and time pressure all affect performance.
- Language. Verbal items measure something different for a test-taker reading in a second language.
- The reference group. A score is relative to whoever the test was normed on. Change the comparison group and the number changes.
Scores on standardised tests have also drifted upward across generations, which is why norms are periodically recalculated. A score from an old norming set is not directly comparable to a recent one.
What the number does not cover
A reasoning score describes performance on a particular set of puzzle types under particular conditions. It is silent on motivation, domain knowledge, creativity, judgement under uncertainty, and the accumulated practice that most real work depends on. Treating it as a summary of a person is a category error, not a cautious caveat.
Reading your own result
A practical way to use a score is to ask what it was measuring and what it was compared against, and then to treat it as one noisy observation rather than a verdict. If the same test gives you a noticeably different number next month, that is the measurement error being visible, not your reasoning changing.
The five-question preview on this site deliberately does not produce a score on this scale: five items cannot support a percentile claim, so it reports how many you answered correctly and nothing more. The methodology page sets out what the longer assessment in the app does, including the fact that its mapping currently rests on published test-score distributions rather than a calibration sample of our own.
If you would rather practise than be measured, the puzzle guides explain solving methods, and the free reasoning preview gives an explanation for every answer.