Why You Can Only Hold a Few Things in Mind at Once — and It's Closer to 4, Not 7

If you've ever heard the phrase "the magic number seven," it comes from a 1956 paper by George Miller called <a href="https://psychclassics.yorku.ca/Miller/" rel="nofollow noreferrer" target="_blank">"The Magical Number Seven, Plus or Minus Two"</a>. It opens with one of the most memorable lines in psychology: "My problem is that I have been persecuted by an integer." Miller had noticed that across very different experiments — recalling digits, comparing tones, identifying tastes — the human capacity to hold separate items in mind kept coming out to roughly seven.

That paper became one of the most cited works in cognitive science. It is also, in the specific form most people remember it, partly wrong.

What Miller actually meant — and what got lost

Miller's careful version of the claim was that short-term memory capacity for unrelated items, with no opportunity to combine them into larger units, was about seven plus or minus two. He was deliberate about the caveats. The number depended on what kind of items, how distinct they were, and whether you could compress them.

In popularization, the caveats fell off. "Seven, plus or minus two" became a clean rule of thumb cited in design guides, productivity books, and undergraduate textbooks, with much less attention paid to what counted as an item.

That ambiguity matters, because items are not stable. The digits "1, 7, 7, 6" might be four items if you read them as separate digits, or one item ("the year of American independence") if you happen to know your U.S. history. Your brain is constantly compressing groups of related things into single chunks before the rest of cognition gets to see them. The number you can hold depends as much on what you already know as on raw capacity.

The revised number

In 2001, the cognitive psychologist Nelson Cowan published <a href="https://www.cambridge.org/core/journals/behavioral-and-brain-sciences/article/abs/magical-number-4-in-shortterm-memory-a-reconsideration-of-mental-storage-capacity/44023F1147D4A1D44BABF089FFE49850" rel="nofollow noreferrer" target="_blank">"The magical number 4 in short-term memory"</a> — a careful review of decades of experiments designed specifically to prevent chunking and rehearsal. When you take those tricks away, the real capacity turns out to be closer to four items, not seven.

This is not a small revision. It is the difference between believing you can comfortably juggle seven distinct things in your head and recognizing that, without aids, you can really only juggle three or four. The reason seven shows up so often in everyday life is that everyday tasks let you compress.

Cowan's number has become the consensus in working memory research over the last two decades. The seven still gets quoted because it is older, simpler, and famous, but the four is closer to what the careful experiments actually find.

Working memory is not one bucket

The other thing that has changed since Miller is the model of what working memory is. Alan Baddeley's <a href="https://www.nature.com/articles/nrn1201" rel="nofollow noreferrer" target="_blank">working memory model</a> — refined over decades — describes working memory not as a single store but as a small set of specialized subsystems: one for verbal/phonological information (the inner voice you use to keep a phone number alive), one for visual/spatial information (the mental sketchpad), and a "central executive" that allocates attention.

This is why holding a number in your head and visually tracking something can both happen at once with less interference than two verbal tasks would have. The systems are partly independent. The four-item ceiling applies within each specialized store, not across them collectively. This is also why "multitasking" is so much harder than it feels — two tasks that compete for the same subsystem destroy each other quickly, even if you experience yourself as handling both.

What this means in practice

Working memory limits show up everywhere once you start noticing them. A few that I have found genuinely useful to keep in mind:

Writing complex code is so tiring because it pegs the executive system. Holding the state of a function, the shape of the data, the contract of an API, and the variable you're tracking all at once — that is a lot of independent items competing for the same constrained resource. The reason a senior engineer often writes simpler code than a junior one is partly that the senior has learned the cost of asking their reader's working memory to hold too much at once. Good code, like good writing, lowers the working-memory load on the next person who reads it.

User interfaces that look "minimal" are not aesthetic preferences. They are accommodations to working memory. A menu with four well-grouped options is genuinely easier to use than the same options in a flat list of twelve, because the smaller set fits inside Cowan's four. Almost every "less is more" claim in UX traces back, at some level, to this.

Chunking is the only real way to increase apparent capacity. You cannot meaningfully expand the underlying limit through brain training apps, no matter what they advertise. What you can do is build domain knowledge through deliberate practice so that what previously felt like ten separate items becomes, with familiarity, two chunks. This is most of what expertise looks like under the hood: an expert holds the same things in mind as a beginner, but the expert's items are bigger.

External memory is not cheating; it is the architecture you are supposed to use. Lists, notes, calendars, code editors, project management tools — all of these exist because human working memory is small and easily overrun. Trying to "just remember" things that could be written down is a form of underusing the tools civilization has built around the actual constraint — and it quietly drains the decision-making capacity that degrades after a long sequence of choices.

A small experiment you can do right now

Open your phone keypad. Look at a 10-digit number you don't know. Look away.

If you can repeat it back exactly while doing nothing else, that is your phonological loop running at near its limit, holding three or four chunks of digits.

Now try the same thing while also keeping in mind a short list of unrelated words. The phone number will start to slip almost immediately. You are watching two pieces of the verbal working memory system compete for the same constrained space. This is the four-item limit, in real time, on a system you use all day.

The honest summary

Working memory is small. It is closer to four items than seven for genuinely unrelated material. It is specialized — verbal, visual, and spatial subsystems compete differently than people often assume. And it is the constraint behind a lot of what we describe as "complexity" — in code, in design, in language, in instructions.

Knowing the limit doesn't expand it. But it does explain a lot of friction in everyday work, and points pretty directly at the right responses: chunk what you can, externalize what you can't, and stop expecting yourself to do something the underlying hardware was never designed for.

If you have come across a recent working-memory result that updates any of this, or have your own example of running into the four-item ceiling, email <a href="mailto:hello@curiospark.org">hello@curiospark.org</a>.