Read this phone number once, then look away and try to recall it: 4-7-2-9-1-8-5-3-6. If you're like most people, you got the first five or six, then the tail end dissolved. You didn't fail — you hit a limit that's baked into your cognitive architecture.
That limit is called working memory capacity, and for decades it was summarized by one of the most famous numbers in all of psychology: 7, plus or minus 2.
Miller's Law: the magical number seven
In 1956, psychologist George Miller published a paper that would become one of the most cited in the field. His conclusion was disarmingly simple: humans can hold roughly 5 to 9 items in short-term memory at any given moment. Seven is the sweet spot. Push past nine and something drops out to make room.
Miller called this the "span of absolute judgment" — the number of distinct chunks of information your working memory can actively manipulate at once. It applies to digits, words, colors, musical notes, and almost every other type of discrete item you might try to juggle mentally.
Later research tightened the estimate. Cognitive scientist Nelson Cowan revisited Miller's data in 2001 and argued the true capacity is closer to 4 chunks, not 7 — the higher number was inflated because people were unconsciously grouping items. But the core insight held: working memory is sharply limited, and the limit is real.
The original study
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. Miller synthesized findings from multiple experiments on absolute judgment and immediate memory, demonstrating a consistent capacity ceiling across stimulus types. The paper remains one of the most cited works in cognitive psychology.
Chunking: how experts cheat the limit
If working memory caps at 7 items, how does a chess grandmaster visualize 20 moves ahead? How does a skilled musician sight-read a page of notes? The answer is chunking — the process of compressing multiple items into a single meaningful unit.
A phone number is a perfect example. "4729185" is seven random digits — that's the ceiling. But "472-918-5" is three chunks: an area code, an exchange, and a number. Same information, far less load on working memory. The digits haven't changed; what changed is how your brain packages them.
Experts do this automatically with domain knowledge. A chess master doesn't see 32 individual pieces — they see "a Sicilian defense with queenside castling." A basketball coach doesn't see 10 players — they see "pick-and-roll with weak-side spread." Years of practice build a library of chunks so rich that experts can hold far more information than novices, despite identical memory capacity.
The practical implication: when you're learning something complex, you're not just acquiring facts. You're building chunks that reduce the working memory burden of using those facts. That's why fluency feels effortless — the chunks have compressed what was once overwhelming into single cognitive units.
The Atkinson-Shiffrin model
Working memory doesn't exist in isolation. It's one stage in a broader memory architecture described by Richard Atkinson and Richard Shiffrin in 1968 — often called the modal model of memory.
The model has three stages: sensory memory (a brief, automatic buffer for raw perceptual input — lasts milliseconds to seconds), short-term memory (the conscious workspace where working memory lives — lasts seconds to minutes, capacity-limited), and long-term memory (effectively unlimited storage, potentially permanent).
The critical insight is that transfer between stages isn't automatic. Information passes from sensory memory to short-term memory through attention — you have to notice it. It passes from short-term to long-term through rehearsal and encoding — you have to process it meaningfully. This is why cramming works poorly: you're cycling information through short-term memory without encoding it deeply enough to reach long-term storage.
Three real-world examples
Why grocery lists exist. Even a modest shopping trip — bread, milk, eggs, pasta, olive oil, garlic, onions, tomatoes — hits the capacity ceiling. Writing it down isn't laziness; it's offloading working memory so you can allocate cognitive resources to navigating the store instead.
Why lecture slides use bullet points. When a slide contains a paragraph of dense text, working memory fills up processing the words before you can think about their meaning. Bullet points chunk the information, reducing the cognitive load to something manageable. Slides that ignore this make audiences slower, not more informed.
Why passwords are hard to remember. "Xk7#mP9!" is eight random characters — near the ceiling, no chunks, no meaning. "correct-horse-battery-staple" is four common words — far more characters, but far fewer cognitive units. The second password is more secure and easier to remember, because it respects how working memory actually works.