← All guides

How much a head can hold

The buffer is tiny, the range is narrow, and the only real upgrade ever found is cheating in a way psychologists approve of.

Updated August 31, 2026

Short-term memory is the most measured faculty in psychology, and its headline result has embarrassed intuition for over a century: the buffer is tiny. Not tiny for some people and vast for others, either; the range across healthy adults is startlingly narrow. What varies between people, and what the four memory tests here actually surface, is mostly not the size of the container. It is what you manage to put in each slot.

Seven, plus or minus two, minus three more

George Miller’s 1956 paper “The Magical Number Seven, Plus or Minus Two” gave the figure everyone quotes: people hold about seven items, whether digits, letters or words. Modern work sharpened the claim and shrank it. When experimenters prevent grouping and rehearsal, the underlying capacity comes out closer to four chunks (Nelson Cowan’s estimate, now the standard one). Seven digits was never seven slots; it was four slots holding groups.

That word, chunk, is the whole game. A chunk is whatever your memory treats as one thing: a digit, a syllable, a phone-number triplet, a chess position if you are a grandmaster. The number memory test looks like it measures digits; from level eight upward it really measures how well you group digits into threes and fours, which is why reading “482 915 37” beats reading “48291537” every time. The famous training studies agree: an undergraduate once pushed his digit span from 7 to 79 over two years, and his letter span stayed stubbornly ordinary. He had not grown the container; he had built an encyclopedia of running times and mapped digits onto it, four digits per chunk.

Order costs extra

Holding items is one job; holding their order is another, and order is the more fragile cargo. Psychology’s instrument for spatial order is the Corsi block task, from 1972: an examiner taps wooden blocks in sequence and the patient taps them back. Corsi spans run about one to two items shorter than digit spans, and the sequence memory test is that task in electric form, with a merciful design twist: the sequence replays in full at every level, so each round rehearses everything you hold and adds exactly one step. The players who go deep are not holding fourteen tiles; they are holding three or four shapes, an L, a zigzag, a bounce, and walking them.

Pictures are the exception

Strip the order requirement away and show the pattern all at once, and spans jump. On the visual memory test most people comfortably hold ten or more tiles, well above any digit span, because a pattern arriving whole gets encoded as a shape, one chunk carrying many tiles for free. The design has to fight this economy to stay hard: the grid grows so the lit tiles never pass half of it (or memorizing the gaps becomes the easier game), and the exposure is pinned to one second so the shape must be taken in, not studied.

The chimpanzee who wins

In 2007, Sana Inoue and Tetsuro Matsuzawa at Kyoto University’s Primate Research Institute published the result that named the chimp test. Young chimpanzees, trained to touch numerals in ascending order, could hold the positions of digits flashed on a screen for a fraction of a second, and did it better than the adult humans tested on the same rig. Ayumu, the best of them, kept his accuracy at exposures around a fifth of a second, a glance no human matches.

The leading interpretation is a trade, not a miracle: chimpanzee working memory keeps an immediate, photographic grip on the visual scene, and the hypothesis is that human ancestors gave some of that grip away as language took over the machinery. We chunk, narrate and compress; Ayumu simply sees. The online version you can play here removes the flash, letting you study the board as long as you like, which is why humans clear it at all. Study time is the human subsidy.

What moves a span, and what does not

Sleep moves it: working memory is the first faculty a short night degrades, which is why every memory test here gives a lapse some mercy (a second try, a strike budget) rather than ending the run on one blink. Age moves it gently. Stress and interruption move it a lot, for the duration of the stress. Training moves the strategy, not the container: after decades of study, gains from memory training transfer to the trained task and its close neighbors, and the four-chunk buffer stays four chunks.

So treat the numbers the way the site does. The Method page declares each test’s curve, which today still leans on published histograms while this site’s own runs accumulate. Your span against the world’s is an estimate; your span against your own last month, played rested on the same device, is a measurement, and grouping better is the one lever that genuinely raises it.