The Magic Number Was Never Seven

Most people will tell you the magic number is seven. That is a simplified version of what George Miller wrote in 1956, and the fuller story is more useful for anyone who solves puzzles.

What Miller actually wrote in 1956

Miller was a cognitive psychologist at Harvard. His paper, "The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information," appeared in Psychological Review. It is often read as saying the average person can hold 7 ± 2 items in short-term memory. The idea is sometimes called Miller's law.

That reading leaves out a lot. According to one summary, Miller used the number partly as a rhetorical device, and later memory research has refined how it is interpreted. Seven was a catchy frame, not a constant stamped on your skull. The more durable idea is that working memory is limited, and that the unit worth counting is the chunk.

Why the limit is about chunks, not items

A chunk is a meaningful unit your mind treats as one thing. What counts as a chunk depends on what you already know. "FBI" may be one chunk for a native English speaker. For someone who has never met the acronym, it is three: F, B, I. Same letters, different load.

Chess is the showy example. A popular account says masters hold 20 or more pieces in roughly five chunks. I could not verify that figure independently, so read it as an illustration. The underlying point is widely made: expertise looks a lot like better chunking. Masters do not need bigger memories. They carry patterns, and a pattern is cheap to hold.

The numbers behind the claim

Miller began with absolute judgment, such as naming a tone from a set of tones. In one-dimensional tasks, people are nearly perfect up to five or six stimuli, then slip as the set grows. That works out to roughly 2 to 3 bits of information, or the ability to tell four to eight alternatives apart.

Memory span is a separate test. It is the longest list you can repeat back in order on half of your attempts. Miller observed about seven items for young adults. Here is the odd part. Span stays about the same even when items carry very different amounts of information: roughly 1 bit for a binary digit, 3.32 bits for a decimal digit, and about 10 bits for a word. If bits were the limit, span should shrink as items get richer. It does not. Miller treated the match with absolute judgment as a coincidence, and counting items turns out to be a poor ruler.

Two more claims circulate, both from a popular secondary account. One says Nelson Cowan's 2001 reanalysis puts the real limit nearer four chunks. The other says one person stretched digit span from 7 to 79 by recoding. I could not read Cowan's paper or find study details for the 79-digit case, so treat both as reported, not confirmed.

Try it: the same digits, read two ways

Here is a quick exercise. Read this string once, cover it, and try to write it from memory:

149217761945

That is twelve digits, which is a lot to hold one at a time. Now try it again, grouped:

1. 1492
2. 1776
3. 1945

If those years mean something to you, the string collapses into three familiar pieces. If they mean nothing, you will feel the difference when you try a string of your own. Ask a friend to make one up, or group a phone number your own way. The digits do not change. Your grouping does, and the load follows.

How recoding works in practice

Recoding is the move you just made. You take small items and fold them into a larger unit that already means something. The secondary account describes recoding as a way to make chunks larger and so raise effective span. It also describes expertise as better chunking, which matches the chess story.

The same account calls the design rule "menus need seven items" a myth. A visible menu is a recognition task, not a recall task, so it sits outside the limit Miller measured. Treat that as one author's interpretation. Even so, it is a useful question to ask yourself while you solve: am I recalling this, or just recognizing it?

A few habits follow naturally. Split long numbers into groups. Give clues short, meaningful labels. Rehearse a pattern until it feels like one thing. None of the sources I had offers tested guidance on training chunking, or on whether brain-training games raise working memory over time. So take these as ideas to try on yourself, not proven methods.

A daily puzzle makes a decent testing ground. Notice how you group a sequence, a grid, or a stack of clues. Notice when a cluster of cells becomes a single shape in your head. That moment is chunking at work.