Miller's law
Working memory holds a small number of chunks, closer to four than to seven.
Miller observed that span tasks and absolute judgement tasks both landed near seven items, and proposed a limit measured in chunks rather than raw items. Later work that blocks rehearsal and grouping puts the real capacity near four. Miller himself treated the number as a coincidence worth investigating rather than a design constant.
The number, before and after revision
How it shows up in software
It is invoked to cap menu items, tab counts and form steps at seven, which is the wrong reading twice over. The number is lower than seven, and it applies to items held in mind, not to items visible on screen. A twelve item menu you can see is not a memory task; a three step flow that asks you to carry a code between steps is.
Using it well
- Apply the limit to what people must hold in their head between screens, not to what is visible in front of them.
- Keep anything needed later on screen: the order summary during payment, the selected filters above the results.
- When a person must carry information between contexts, such as a verification code, reduce it to one chunk or autofill it.
- Chunk long sets rather than truncating them. Grouping twelve links under three headings respects the limit without removing anything.
Where it turns manipulative
- Deleting navigation items to satisfy a rule of seven, which removes function to honour a number that was never a design rule.
- Quoting seven plus or minus two in a design review as though it settles a layout argument. Cowan revised it to about four, and neither number governs visible lists.
- Splitting a form into many short steps to keep each under seven fields, producing more screens, more state to carry and a longer flow.
Where you have seen it
Stripe Checkout
The order summary stays visible beside the payment fields, so nothing about the purchase is held in memory.
iOS
One-time passcodes from messages are offered directly above the keyboard, removing the need to carry a code between apps.
Amazon
Navigation runs to dozens of departments, grouped and visible, rather than being cut to a memorable handful.
What the research says
- Miller, 1956Mixed evidence
Reviewed span and absolute judgement experiments that clustered around seven plus or minus two, and argued the limit applies to chunks, so recoding raises effective capacity.
The paper is a synthesis with a semi-humorous framing, not a single controlled study. Miller called the recurring number a coincidence he was being persecuted by.
- Cowan, 2001Well evidenced
Reviewed studies that prevent rehearsal and chunking and concluded pure working memory capacity is about four chunks in young adults.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.