By Eric, BlinkBench Founder · Last reviewed: July 2026
“The average memory span is seven, plus or minus two.” It's one of the most repeated facts in casual psychology, usually traced back to a 1956 paper with that number in its title. It has also outrun what any single study can actually support, and the paper this site cites for its own digit-span comparison happens to contain a clean demonstration of why.
Two samples, one paper, a full digit apart
The study behind number memory's reference range ran two experiments. The smaller one tested 30 adults aged 18–46 and found a mean forward span of 7.36 digits (SD 0.88), with individual spans ranging from 5.7 to 9.0. The larger one tested 763 adults aged 18–65 and found a mean of 6.35 digits (SD 1.15).
Same task, same lab, same one-digit-per-second presentation, same recall procedure. The means are 1.01 digits apart — a full item, which on a scale that moves in whole digits is not a rounding difference. Nothing about digit span itself changed between the two experiments. What changed was the sample: 30 people versus 763, a narrower age band versus a wider one. That alone was enough to move the average by almost a full unit of the thing being measured.
If a single published paper can report two means a digit apart from two samples of the same task, then “the average is seven” was never a fact with the precision the sentence implies. It's a rough center of a number that moves with who you ask.
And that's before the task changes
Digit span isn't the only span number in circulation, and the others aren't interchangeable with it or each other. Sequence memory's reference task — the Corsi block-tapping test — measures spatial rather than verbal memory, and the largest recent normative sample for it, 340 adults aged 21–89, found a mean forward span of only 4.97 blocks (SD 1.03) — more than two whole items below either digit-span figure above.
That is not spatial memory being weaker than verbal memory in some general sense. It's two different tasks, built from different material, standardized by different researchers, producing numbers that were never on the same scale to begin with. Quoting “my memory span is seven” without saying which task produced it is already underspecified; assuming it transfers to a spatial task compounds the error.
The stopping rule is doing more work than it looks like
There's a second, quieter source of drift: how a test decides you've hit your ceiling. Every span test on this site — number memory, sequence memory — ends a run after 2 misses at the same length, not one. That rule is shared code (lib/span.js), not a per-test tuning knob, for a specific reason: a single miss is weak evidence that a length is genuinely beyond someone. It could be a mistyped digit, a mis-tapped block, one lapse of attention — the reference studies cited on both pages use the same reasoning, requiring more than one failure before calling a length the ceiling.
Loosen that rule to one strike and scores drop, because ordinary slips start ending runs that would otherwise have continued. Loosen it further — three or four attempts per length — and scores climb, because sheer repetition eventually produces a lucky pass at a length someone can't reliably hold. Two studies measuring “the same” span with different stopping rules are measuring related but distinct things, and the rule is rarely the first thing quoted alongside the number.
What this means for reading your own score
Not that span numbers are meaningless — the tests measure something real and repeatable, which is why this site cites them at all. It means a specific number like “7” carries less precision than the folklore version of it suggests. The honest unit isn't a single average — it's a range, tied to a specific task, a specific sample, and a specific stopping rule, all of which this site states plainly next to the figure rather than compressing into one folklore-sized number. That is also why verbal memory doesn't report a span at all — a streak of correct answers on that task can be inflated just by answering “seen” more often, which is a different problem again, and the page uses a different measure (d′) built specifically to resist it.