Start with the range, not the number
A reported IQ is an estimate carrying error, and the honest unit is an interval. A 95% confidence interval is the observed score plus and minus 1.96 × the standard error of measurement — with an SEM of 3, roughly six points in each direction.
This is not a technicality. It decides whether two scores mean anything different. A person at 128 and a person at 132 on the same instrument are not reliably distinguishable; one clears the Mensa threshold and one does not, which is a fact about where a line was drawn rather than a fact about their reasoning.
What 100 is, and what it is not
Modern IQ is a deviation score. The population mean is pinned at 100 and the standard deviation at 15, then performance is expressed as a position on that curve. Nothing is divided by anyone's age; the ratio method that produced the phrase “intelligence quotient” was abandoned decades ago and the name outlived it.
So 100 is not a pass mark and not 100% of anything. It is the middle. A score of 115 sits one standard deviation up, around the 84th percentile.
Bands and percentiles
| Score | Percentile | Conventional label | Frequency |
|---|---|---|---|
| 145+ | 99.9th | Highly gifted | 1 in 1,000 |
| 130–144 | 98th–99.8th | Very superior | 1 in 50 |
| 120–129 | 91st–97th | Superior | 1 in 12 |
| 110–119 | 75th–90th | High average | 1 in 6 |
| 90–109 | 25th–74th | Average | 1 in 2 |
| 80–89 | 9th–24th | Low average | 1 in 6 |
| 70–79 | 2nd–8th | Borderline | 1 in 12 |
| Below 70 | Under 2nd | Extremely low | 1 in 50 |
Labels are clinical shorthand describing test performance. Half the adult population falls between 90 and 109, which is why that row is so crowded and why small movements inside it mean very little. How our own bands are derived, and how wide the error band around them is, sits in the methodology notes.
Percentile does not mean what most people think
A percentile of 84 does not mean you are “smarter than 84% of people”. It means 84% of the comparison group scored lower on this test, on this occasion. Change the instrument, the sample or the day and the figure moves.
The comparison group matters most and gets examined least. Clinical norms are matched to census demographics. Online norms are drawn from people who chose to take an online IQ test, which skews younger, more educated and more curious about the topic than the population at large. A percentile from a web screener describes your position among that self-selected crowd.
1. Treating a point estimate as exact when the instrument reports an interval.
2. Comparing scores across instruments with different standard deviations.
3. Reading a percentile as a statement about the whole population when it describes a specific sample.
The same position, three different numbers
Not every test uses SD 15, so an identical percentile carries different figures. This is the source of most confusion about quoted scores.
| Instrument | SD | 84th pct | 98th pct | 99.9th pct |
|---|---|---|---|---|
| Wechsler (WAIS, WISC) | 15 | 115 | 130 | 146 |
| Stanford–Binet 5 | 16 | 116 | 132 | 149 |
| Cattell III B | 24 | 124 | 148 | 174 |
A 148 on Cattell and a 130 on Wechsler are the same place in the distribution. Whenever someone quotes a number, the useful question is which scale it came from.
Why your score does not drop as you age
IQ scores are age-normed: you are compared against people in your own age band, not against twenty-two-year-olds. Average IQ therefore stays near 100 across the lifespan by construction, even though raw performance on some tasks changes a great deal.
The underlying abilities do move. Raymond Cattell drew the distinction in 1963 between fluid intelligence (Gf), reasoning about novel problems, and crystallised intelligence (Gc), accumulated knowledge and vocabulary. Horn and Cattell showed in 1967 that the two follow opposite trajectories: Gf peaks in early adulthood and slides gradually afterwards, while Gc rises through middle age and holds. Salthouse put a figure on the fluid decline — on the order of 0.02 standard deviations a year from about age 20, accelerating somewhat after the mid-fifties.
There is no single cognitive prime
The neatest evidence against the idea that you peak at twenty and coast comes from Hartshorne and Germine, published in Psychological Science in 2015. They combined 48,537 online participants with normative data from standardised IQ and memory tests and looked at when performance peaks across roughly thirty tasks. The peaks were scattered:
| Ability | Approximate peak |
|---|---|
| Processing speed | 18–19 |
| Short-term memory | Mid-twenties |
| Working memory | Late twenties to early thirties |
| Recognising and remembering faces | 30–34 |
| Reading emotion in others | Forties to around fifty |
| Vocabulary | Late fifties to sixties |
A single composite score flattens all of that into one figure. Two people with the same IQ at forty may have arrived there by very different routes, and the composite hides which.
Above the Mensa line
Mensa admits at the 98th percentile, which is roughly 130 on the Wechsler scale, 132 on Stanford–Binet 5, and 148 on the Cattell III B used by British Mensa. Only supervised results count; American Mensa charges $60 for its own admission test and sets a minimum age of 14.
Societies exist with higher thresholds — the 99th percentile and above — and they get progressively stranger as the cutoff rises, because no instrument can reliably distinguish people that far into the tail. Above roughly the 99.9th percentile the measurement error exceeds the differences being claimed.
What the number does not predict
Group-level correlations between IQ and educational or occupational outcomes exist and are moderate. Moderate correlations at group level say very little about any one person, which is the step most commentary skips.
An IQ score does not measure creativity, judgement, conscientiousness, motivation, emotional skill or integrity, and it cannot identify a learning disability, ADHD, autism or giftedness. Those need a full assessment in which a cognitive test is one component among several.
This is also why you will not find celebrity IQ tables or national rankings anywhere on this site — the reasoning is in what we refuse to publish, and the usage terms list the decisions a result here cannot support.
Sources
- Cattell, R. B. (1963). Theory of fluid and crystallized intelligence: a critical experiment. Journal of Educational Psychology, 54, 1–22. The Gf/Gc distinction.
- Horn, J. L. & Cattell, R. B. (1967). Age differences in fluid and crystallized intelligence. Acta Psychologica, 26, 107–129. The opposite lifespan trajectories.
- Salthouse, T. A. (2009). When does age-related cognitive decline begin? Neurobiology of Aging, 30, 507–514.
- Hartshorne, J. K. & Germine, L. T. (2015). When does cognitive functioning peak? The asynchronous rise and fall of different cognitive abilities across the life span. Psychological Science, 26, 433–443. n = 48,537.
- American Mensa — published qualifying criteria and supervised testing fee. Checked 22 July 2026.
- Pearson Clinical Assessment — WAIS-5 product documentation, released 2024. Checked 22 July 2026.
Checked against the sources above on 22 July 2026. Found an error? Send it over — corrections are read before anything else.
If your score surprised you
Take the interval seriously before taking the number seriously, and consider the conditions: time of day, interruptions, whether it was a first attempt. A single unsupervised screener taken once is weak evidence about anything, and the right response to a surprising result is usually curiosity rather than conclusion. If it matters — for a diagnosis, an accommodation, an application — a supervised assessment is the only thing that will carry weight.
Think a band on this page is described wrongly, or a figure has moved? Send it over — corrections are read before anything else. If you would rather know what happens to your data before you take the test, the privacy inventory and the note on local storage answer that in full.