By Eric, BlinkBench Founder · Last reviewed: September 2026
Picture a phone number you already know by heart — ten digits, no hesitation. Now picture typing back ten random digits shown one at a time on Number Memory. The published reference figure this site cites for that exact task puts the average forward span at 6.35 digits in a sample of 763 adults, and 7.36 digits in a separate, smaller sample — several digits short of ten either way. The gap between a phone number you already know and a random string that long isn't about how many digits either one contains. It's about whether there's a way to turn many digits into a few things worth remembering.
That gap has a name in cognitive psychology: chunking. It's the same idea behind one of the field's most famous papers, and behind an experiment that took an ordinary undergraduate's digit span from 7 to nearly 80.
Capacity measured in chunks, not items
In 1956, the psychologist George Miller published "The Magical Number Seven, Plus or Minus Two," opening it with a line that's stuck around almost as long as the number itself: "My problem is that I have been persecuted by an integer. For seven years this number has followed me around, has intruded in my most private data, and has assaulted me from the pages of our most public journals."
The paper's real argument is easy to miss under that number. Miller was drawing a distinction between two different limits on how we process information. One is what he called channel capacity — "the greatest amount of information that he can give us about the stimulus on the basis of an absolute judgment." The other is the limit on immediate memory, and Miller was explicit that it isn't the same kind of limit: "the span of absolute judgment and the span of immediate memory are quite different kinds of limitations that are imposed on our ability to process information. Absolute judgment is limited by the amount of information. Immediate memory is limited by the number of items."
Number of items, not amount of information — and an "item," Miller argued, doesn't have to be a single digit. He was candid that the concept was fuzzy even as he introduced it: "the contrast of the terms bit and chunk also serves to highlight the fact that we are not very definite about what constitutes a chunk of information." But he was specific about the mechanism and its payoff: "we must recognize the importance of grouping or organizing the input sequence into units or chunks. Since the memory span is a fixed number of chunks, we can increase the number of bits of information that it contains simply by building larger and larger chunks, each chunk containing more information than before." A phone number is ten digits recoded into a handful of chunks — an area code, an exchange, a line number. A random string offers no such recoding for free.
What chunking looks like taken to an extreme
Miller's paper was theoretical. A 1980 study gave the idea a face. Ericsson, Chase and Faloon trained a single college student, described elsewhere as "a college student of average intelligence," on a digit-span task over many months. The original paper sits behind a paywall at Science, and a Defense Technical Information Center copy that normally mirrors it was down for maintenance when this guide was researched — so the figures below are drawn from Wikipedia's summary of the study instead. According to that summary, "SF began the experiment with a normal span of seven digits," and "by the end of the experiment, his digit span had grown to 80 numbers" — an increase the same summary describes as happening "within 20 months or after at least 230 hours" of practice.
S.F. didn't get faster or more attentive. He built a system. He was a long-distance runner, and the same source reports that "chunking strings of digits into race times increased his digit span" — turning an arbitrary run of digits into something with meaning he already had a place to store. Even so, the chunks themselves stayed small: the summary notes he "'chunked' only three to four digits at once," stacking multiple small, meaningful groups rather than holding one enormous one. And the skill didn't generalize. The same source reports a check on whether S.F.'s underlying memory capacity had simply grown: "if his short-term memory capacity increased, then he would have shown a greater capacity for the alphabets; he did not." Nearly 80 digits, and an entirely ordinary span for letters. What had grown was a digit-specific retrieval system, not memory itself.
What this means for reading your own score
Number Memory's own page explains why digits are shown one at a time rather than all at once: shown all at once, "you can re-read it, chunk it into pairs, and rehearse the chunks" — and the reference figures above weren't collected that way. That's a real constraint, and it does close off that re-scanning version of chunking.
It doesn't close off the version Miller described or the version S.F. spent 230 hours building. Both of those happen as digits arrive, not by looking back at ones already shown — recoding an incoming digit or pair into something already meaningful before the next one appears. Nothing about a one-second cadence prevents that. Number Memory's own page states that each digit is "drawn uniformly at random from 0–9, independently each time," so no built-in phone-number-style structure is waiting to be found — but a test-taker who happens to notice a birth year, a familiar price, or a number they already track for other reasons is doing exactly what S.F. did, just without 230 hours behind it.
That's worth knowing before comparing a number-memory score against someone else's, or against the cited reference range. Two people can differ in this test for reasons that have nothing to do with which one holds more in working memory: one may simply have more prior material — dates, prices, sports statistics, running times — sitting ready to absorb an incoming digit or two. That's a real skill, and S.F.'s case shows it can be built deliberately. It's just a different skill from the one "digit span" is usually assumed to measure on its own.
Frequently asked questions
Does this mean a high number-memory score isn't really about memory?
It's about memory plus whatever chunking a person manages to apply in the moment, and the test has no way to separate the two. A string that happens to contain a familiar date or a number you already track gets an assist that a purely random string, to someone with no matching background knowledge, doesn't get. Both people are still using memory — one of them just has less raw material to hold because part of the string arrived pre-packaged.
How many digits fit in one chunk?
In S.F.'s case, according to the secondary summary cited above, three to four digits at a time — he stacked several such chunks rather than holding one large one. There's no reason to assume that number is fixed for everyone or every kind of chunk; it depends on how compactly a person can recode a given group into something already familiar.
Can I train myself to chunk better on this test?
S.F.'s case says the underlying trick — recoding groups of digits into things you already know, like dates or times — can be built with enough practice, and it didn't take unusual raw talent: he's described as "a college student of average intelligence" going in. But the same case also shows the skill stayed narrow: it didn't carry over to letters. Practicing digit chunking would likely raise a number-memory score without telling you anything new about memory more generally, which is part of why this site doesn't frame retaking any test here as "training" — see the companion guide on practice and this site's tests.
Why doesn't the test just give everyone the same chunking opportunities, like grouping digits into pairs?
Because the digits are drawn independently and at random, there's no fixed structure to group by design — imposing one (say, always displaying pairs) would hand every test-taker the same partial chunk for free and change what the score reflects for all of them at once, in a way no cited reference range was measured under. What chunk, if any, a given string offers is left to chance and to what each person already happens to know.
Sources
- Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97.
- Woods, D. L., Kishiyama, M. M., Yund, E. W., Herron, T. J., Edwards, B., Poliva, O., Hink, R. F., & Reed, B. (2011). Improving digit span assessment of short-term verbal memory. Journal of Clinical and Experimental Neuropsychology, 33(1), 101–111.
- Wikipedia. Chunking (psychology) — summary of Ericsson, K. A., Chase, W. G., & Faloon, S. (1980), Acquisition of a memory skill, Science, 208(4448), 1181-1182.
- Wikipedia. Mnemonist — describes subject S.F. from Ericsson, Chase & Faloon (1980).