What is a good IQ score?

110 is good in a plain sense: about three in four people score below it. It is the start of the high average band. Read the full picture for an IQ of 110.

120 is clearly good. About 1 person in 11 scores this high. Read an IQ of 120 for the percentile and rarity.

130 is the point most people mean by a high IQ. It is two standard deviations above average, about 1 in 44. It is also the Wechsler score American Mensa accepts. See an IQ of 130 and the Mensa eligibility checker for how the line differs between tests.

Yes. A good IQ on one scale is a different number on another. The Wechsler tests and the current Stanford-Binet use a spread of 15 points, an older Stanford-Binet used 16, and the Cattell tests use 24. A Cattell score of 148 and a Wechsler score of 130 describe the same rank. Our guide to the standard deviation shows how to convert between them.

The quality of the test matters even more. A supervised test with a representative norm sample gives a figure you can compare with the table above. Many short online tests have no such norms, so their numbers do not map onto it. How accurate online IQ tests are explains what to look for.

The scale does not change with age. A good IQ for a 10-year-old and for a 40-year-old is the same number, because each is compared with people their own age. Our average IQ by age pages explain how age norms work.

Schooling does move scores. Ritchie and Tucker-Drob's 2018 meta-analysis of 42 datasets found that each extra year of education raised cognitive test scores by about 1 to 5 IQ points. So a score is not a fixed label, and a score taken at 12 can differ from one taken at 30.

Less than the digits imply. Every IQ score has a margin of error. On a full professional test the 95% confidence band is roughly 5 points either side, so a reported 118 is best read as somewhere from about 113 to 123. Two scores a few points apart, or two attempts a few weeks apart, are usually the same result seen through noise.

Sleep, mood, illness and the test setting also matter, and practising the item format helps. For the maths behind the margin, see the standard deviation page, and for the tool, the IQ confidence interval calculator.

There is no scientific cut-off for either word. On the Wechsler labels, every score from 130 upwards is simply "very superior". Mensa admits the top 2%. Intertel takes the top 1%, about 135 on the standard scale, and the Triple Nine Society takes the top 0.1%, about 146. Each society picks its own line, as the Mensa eligibility checker shows.

So a good IQ does not become "genius" at a particular number. Words like gifted and genius are conventions, and in everyday use they also imply achievement that a test score cannot show. The IQ classifications page lists the labels used for each band.

Less than the number suggests, and more than critics claim. IQ is one of the better predictors in psychology. In a meta-analysis of longitudinal studies, Strenze found that intelligence correlated .56 with later education, .43 with occupational level and .20 with income. Those are real relationships, and they also leave most of the variation unexplained. Our page on IQ and income goes through the numbers.

Two other limits apply. Each score has a margin of error of a few points, so 118 and 122 are not meaningfully different. And IQ does not measure persistence, creativity, social skill or judgement, which matter at work and in life.

On the Wechsler labels, 80 to 89 is low average, 70 to 79 is borderline and below 70 is extremely low. About 2% of people score below 70. A diagnosis of intellectual disability needs more than a score: it also requires problems with day-to-day adaptive skills, assessed by a clinician.

A single low result from an unsupervised test is not evidence of anything. Sleep, stress, familiarity with the format and effort all move a score.

The reliable route is a supervised test such as the WAIS, given by a qualified psychologist. Our guide to the WAIS test online explains what it involves. For a first look, an online reasoning test can show how you handle pattern puzzles, as long as you treat the result as an estimate.