Miller's Law

012 min

Which Miller's Law this entry means

Four unrelated rules carry the name Miller's Law. This entry covers the memory one.

Called Miller's Law	Field	What it says
The magical number seven	Psychology	Immediate memory holds about seven chunks. George A. Miller, 1956.
The listening rule	Communication	To understand a speaker, assume the statement is true, then work out what it would have to be true of. Also George A. Miller.
Postnasal deaspiration	Linguistics	A sound change in ancient Greek. Nothing to do with the psychologist.
The Bugzilla rule	Software	A bug discussion tends to drift toward proposing a full redesign. Named by Mike Beltzner after Dave Miller.

The listening rule is the one most often mixed up with the memory finding, because the same person is credited with both. It asks you to hold back your judgement for a moment. Rather than asking whether the speaker is wrong, you ask what situation would make the statement correct, and then ask whether the speaker is in that situation. That is advice about listening. It says nothing about how much anyone can hold in mind.

The evidence behind it is thinner than the 1956 paper's. The wording is usually traced to a 1980 interview Miller gave Elizabeth Hall in Psychology Today. That interview is not online, so most readers get the quotation secondhand. Wikipedia's Miller's law page lists all four senses with the references for each.

Everything after this section is about the memory finding.

022 min

What Miller measured in 1956

Miller first read the paper as an invited address to the Eastern Psychological Association in Philadelphia on 15 April 1955. It was published the next year in Psychological Review, volume 63, pages 81 to 97. He opens it as a complaint.

"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."

George A. Miller, 1956

The paper reviews three separate lines of experiment that each produced a number near seven. Miller's argument is that they are not the same limit.

The span of absolute judgement

A person is given one tone and has to say which pitch it was, with no second tone to compare it against. Miller collected results across pitch, loudness, brightness, saltiness, curvature and position along a line. The scores ranged from 1.6 bits for curvature to 3.9 bits for position in an interval, with a mean of 2.6 bits. A bit is a unit of information: one bit is the answer to one yes-or-no question. Miller converted 2.6 bits to about 6.5 categories a person can tell apart. This is a limit on grading one physical property.

The span of immediate memory

This is the length of list you can repeat back straight after hearing it. Miller cites Hayes, who read lists aloud at one item per second. The span was about nine for binary digits, about seven for decimal digits, and about five for one-syllable English words drawn at random. This is a limit on holding a list.

The span of attention

About six objects taken in at a glance.

Miller then names the error he wants to stop. Treating these three spans as one underlying process, he writes, "is a fundamental mistake". Almost every use of Miller's Law outside psychology means the second span only.

032 min

Why Miller's Law counts chunks, not items

The finding is stated in chunks because the list length changes with the material and the person, while the chunk count stays fairly steady. Miller's own words for the two units are bits and chunks. Absolute judgement holds a constant number of bits. Immediate memory holds a constant number of chunks, and a chunk can carry very different amounts of information depending on what the person already knows.

Take the letters C, A and T. To someone who cannot read English they are three units. To someone who can, they are one. Nothing about that second person's memory is larger. What differs is what has already been learned as a single unit, and that is what Miller's count applies to.

Miller calls the act of building bigger chunks recoding, and the paper contains a demonstration of it. Sidney Smith measured the spans of 20 people for binary digits and for octal digits: about nine binary digits and about seven octal digits. Then he taught them to rewrite binary digits in groups. Two binary digits become one digit from 0 to 3; three binary digits become one digit from 0 to 7. Every group improved. None improved as much as their octal span predicted. Miller's reading was that five or ten minutes of study is not enough for the translation to run automatically.

So Smith drilled himself over a longer period. He could hold 12 octal digits. Once the four-to-one and five-to-one groupings were automatic, those 12 chunks were worth about 40 binary digits, and he could repeat back a 40-digit binary string without error.

This is the part of Miller's Law that transfers. The limit is not a fixed number of things on a screen. It moves with how much of the material the reader has already learned to treat as one unit.

042 min

Miller's Law and the billboard laws

Miller set out the clearest correction himself, in a private reply that is now public. In July 1998 Mark Halpern, an editor, wrote to Miller at Princeton. Halpern's technical writing director had issued a rule that lists and procedures in their documentation must not exceed seven items, or nine at most. Halpern wanted Miller's own words to argue against it.

Miller replied with an older story. Local authorities had used his paper to pass laws limiting how many items a billboard could carry. The same argument was used for a United States law banning billboards within a set distance of highways. Large motel chains benefited, because a traveller already knows what a chain offers. A single independent motel has to state its services on the sign, so an item cap took away its only way to compete. The billboard industry hired someone to tour towns disputing the claim. That man's wife, a psychologist, told him the paper did not say what everyone thought it said, and he traced it back to Miller.

Miller wrote a correction letter to a trade journal at the time. He told Halpern he no longer had a copy and could not recall the journal's name, so the letter itself is not checkable. His summary of it is:

"the point was that 7 was a limit for the discrimination of unidimensional stimuli (pitches, loudness, brightness, etc.) and also a limit for immediate recall, neither of which has anything to do with a person's capacity to comprehend printed text."

George A. Miller to Mark Halpern, 1998

On Halpern's seven-item documentation rule, Miller answered that nothing in his paper warrants asking Moses to discard any of the ten commandments.

052 min

How Miller's Law became a seven-item rule

Edward Tufte published the Miller letters in 2003 alongside his own reading of the paper. His objection is that the experiments were about repeating back unrelated material, which is not what a reader does with a list, a slide or a menu. Tufte writes that the conclusion that only seven items belong on a list "can be sustained only by not reading the paper". He adds that the paper's actual advice points the opposite way. Put information in a context, and memory reaches further.

The reason the rule fails for interfaces is simpler than the debate about the number. A menu is visible while you use it. Nothing has to be held in mind at all. Jakob Nielsen made this point at Nielsen Norman Group in 2009: longer menus are fine, because users do not have to memorise the list. Kate Moran repeated it there in 2016, adding that the useful part of Miller's work for designers is chunking, not counting.

Where the seven-item rule has been tested against a real alternative, it lost. Larson and Czerwinski compared site structures of different width and depth at CHI in 1998. Kath Straub summarised the result for Human Factors International in 2003. Her practical figure is roughly 16 top-level links leading into two or three further levels. She concludes that "moderate breadth affords optimal user performance". Both the very deep structure and the very wide one did worse than the middle.

Three tells that a team is applying the rule rather than the finding:

The number is applied to things the user can see, such as navigation links, table columns, or cards on a page.
The cap is the only reason given. Nobody has asked what the user has to hold in mind between two screens.
Removing an item made the remaining labels vaguer, which costs more than the item did.

062 min

What the real limit in Miller's Law is

Miller's seven has been revised downward by the researchers who followed him, and the revision is about method rather than about people getting worse.

Estimate	Who, when	What it counted
About 7 chunks	Miller, 1956, Psychological Review	Lists repeated back with rehearsal and grouping allowed
5 to 7 chunks	Simon, 1974, Science	Chunk capacity pooled across several experiments
3 to 5 chunks	Cowan, 2001, Behavioral and Brain Sciences	Lists where grouping and rehearsal were blocked
3 to 5 chunks	Cowan, 2010, Current Directions in Psychological Science	Same limit, restated as a central store
About 4, or about 7	Morra, Patella and Muscella, 2024, Journal of Cognition	Depends on presentation time and what is counted as a unit

Herbert Simon pooled results across several experiments in 1974. He reported that "the chunk capacity of short-term memory has been shown to be in the range of five to seven".

Nelson Cowan's 2001 review is the one most often cited against the seven. He argues that Miller's number "was meant more as a rough estimate and a rhetorical device than as a real capacity limit". His own figure is three to five chunks. His method is the interesting part. A capacity limit only becomes visible when the person is stopped from building larger chunks. Cowan set out four conditions under which that happens, including overloading the input and blocking rehearsal. Under those conditions the average is about four.

This is still an open argument, not a settled replacement. Morra, Patella and Muscella reported in the Journal of Cognition in 2024 that both numbers survive, because they measure different things. A long look at the material "yields (artifactually) higher capacity estimates", because the person has time to group it. A brief look does not. Alan Baddeley had revisited the paper in the same journal in 1994, 38 years on. His title was "The magical number seven: still magic after all these years?".

Miller's own verdict on the number was that he suspected "it is only a pernicious, Pythagorean coincidence".

072 min

Miller's Law and a 79-digit memory span

The strongest evidence that Miller's Law limits chunks rather than items comes from one person studied over two years. K. Anders Ericsson, William Chase and Steve Faloon published the result in Science in 1980, volume 208, pages 1181 to 1182.

Their participant, known in the literature as SF, started with an ordinary digit span of seven. He was a long-distance runner, and he began recoding groups of digits into race times he already knew. A group of three or four digits stopped being three or four things and became one thing.

After more than 230 hours of practice his digit span reached 79. The authors report that his performance on other digit tasks matched that of people with lifelong memory training. Their conclusion was that "with an appropriate mnemonic system, there is seemingly no limit to memory performance with practice."

One control makes the case. Norris, Hall and Gathercole reviewed the work in Memory & Cognition in 2019. They note that SF also trained on letter sequences. There, "his span remained at seven across the full training period". He had no running times for letters, so he had no way to build larger units, and the underlying limit reappeared unchanged.

Set this beside Sidney Smith drilling himself to 40 binary digits in Miller's own paper. The two cases differ in one variable. Smith used an arithmetic rule anyone can be taught in minutes. SF used knowledge he had spent years acquiring outside any laboratory. The span reached was roughly twice as high in the second case.

Neither man enlarged his memory. Both changed what one unit meant, and the letter result shows what happens the moment that option is taken away.

082 min

Miller's Law in product, design, and AI work

For a builder, the finding applies wherever something must be carried in mind from one place to another, and rarely anywhere else.

The clearest real case is number formatting. Payment card numbers are printed and typed in groups of four rather than as sixteen digits, and international phone numbers are spaced into groups too. The grouping makes each segment one unit to hold while your eyes move between the card and the field. A checkout that strips the spaces out of a card field removes that help for no gain.

The case that costs the most is a flow that forces a user to hold information across a screen change. One-time codes are the common example. A code sent to another application has to survive a switch away from the page and back. Six digits in two groups of three survives that switch. Six digits run together, or a code the form clears when the user returns, does not. The fix is not fewer digits. It is grouping, plus letting the value be pasted.

For an analyst, the same limit shows up in labelling and evaluation work. Consider a researcher writing a rubric for people scoring model outputs, and giving each rater nine criteria to weigh at once. Raters cannot hold nine, so they settle on two or three and score the rest inconsistently. Splitting the rubric into passes of three, each with a decision at the end, changes what the rater has to hold rather than what the rubric asks for. The rubric is still nine criteria long.

Note which direction the design is pointing. Grouping a card number helps the person. A comparison table built so that no single screen shows two competing prices is using the same limit against them.

092 min

How to use Miller's Law without misusing it

Apply the finding to memory load, and never to item counts on a visible surface.

Ask what has to be held, not what is shown.
If the information is on screen while it is needed, Miller's Law does not apply. Nielsen Norman Group's recognition-over-recall guidance covers that case instead.
Group, do not cut.
Break long strings, forms and instructions into units of three or four. That is the part of the 1956 paper that carries over, and it costs the reader nothing.
Make the units match what the reader already knows.
A grouping that maps onto something familiar becomes one chunk. An arbitrary grouping stays several. This is the difference between Ericsson's participant and an untrained one.
Count the handoffs.
Every point where a user must carry a value between two screens is a place where the limit binds. Reduce the number of handoffs before you reduce the number of digits.
Plan for three or four, not seven.
Cowan's figure is the safer one for anything the user has to keep in mind unaided.
Do not cite the number to win an argument about layout.
Miller answered that question himself in 1998, and Tufte's objection has stood since 2003 without a serious reply.

One honest note. Knowing about the limit does not raise it. Nothing in this literature shows that awareness of Miller's Law improves anyone's recall. What changes performance is the material: better grouping, better labels, and fewer moments where something has to be carried in the head at all.

102 min

Miller's Law vs. nearby concepts

Concept	How it differs from Miller's Law
DesignHick's LawAbout how long a choice takes as the number of options rises. It concerns decision time, not memory. The options are visible throughout.
PsychologyChunkingThe mechanism inside Miller's Law rather than a rival to it. Chunking is the act of grouping; Miller's Law is the claim about how many groups fit.
PsychologyWorking memoryThe broader system, which holds and manipulates information. Miller's Law describes one capacity limit within it, measured by repeating a list back.
Not in the library yetCognitive loadA claim about effort during learning or a task. Miller's Law is a claim about storage capacity at one moment. High load can come from a single hard idea, with no list involved.
Not in the library yetRecognition over recallA design rule that avoids the limit rather than working inside it. It says put the options on screen, so nothing has to be remembered.
Not in the library yetCowan's numberThe same measurement taken under stricter conditions, giving three to five instead of seven. Not a different concept, a corrected value.

The axis that separates most of these is whether the information stays visible. When it does, choice time and scanning rules apply and Miller's Law does not. When it disappears between one moment and the next, Miller's Law is the constraint, and the number to plan around is Cowan's, not Miller's.

?7 questions

Questions people ask

What does Miller's Law actually say?

It says the span of immediate memory is about seven chunks, not seven items. George Miller published it in Psychological Review in 1956. A chunk is any group the mind already treats as one unit, so the number of things a person can hold changes with what they already know.

Is Miller's Law about memory or about communication?

Both, confusingly. George Miller is credited with two separate rules. The 1956 memory finding is the one design and product work means. The communication rule, which says assume a statement is true and work out what it could be true of, is traced to a 1980 Psychology Today interview.

Does Miller's Law mean menus should have seven items?

No. A menu is visible while you use it, so nothing is held in memory. Jakob Nielsen made this point at Nielsen Norman Group in 2009, and Edward Tufte wrote that the seven-item conclusion "can be sustained only by not reading the paper". Miller himself rejected the reading in 1998.

Is the real limit seven or four?

Nearer four when people are stopped from grouping. Nelson Cowan argued in Behavioral and Brain Sciences in 2001 that the limit is three to five chunks, and that Miller's seven came from tasks that allowed rehearsal. The argument is still live: Morra, Patella and Muscella reported in 2024 that both numbers hold, for different measurements.

What is a chunk in Miller's Law?

A chunk is whatever the mind has already learned to treat as one unit. The letters C, A and T are three chunks to a non-reader of English and one chunk to a reader. Miller called the act of building larger chunks recoding, and showed it raising a binary digit span from about nine to about 40.

How do I apply Miller's Law to a product?

Apply it to what a user must carry in their head, not to what is on screen. Group long strings into units of three or four, as payment card numbers are printed. Then count the points where a value must survive a screen change, and reduce those before reducing anything else.

Did George Miller believe the number seven meant something?

He doubted it. He closed the 1956 paper by listing the seven seas, the seven deadly sins and the seven days of the week. He then wrote that he suspected the recurrence was "only a pernicious, Pythagorean coincidence". He also opened the paper by saying he had been "persecuted by an integer".

▶3 videos

Watch

Understanding Miller's Law
Laws of UX: Miller's Law (with examples!)
AM Design
The Magical Number 7 and UX
NNgroup

§15 sources

Sources for Miller's Law

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.
Full text · psychclassics.yorku.ca(opens in a new tab)PDF · labs.la.utexas.edu(opens in a new tab)record · pubmed.ncbi.nlm.nih.gov(opens in a new tab)
Simon, H. A. (1974). How big is a chunk? Science, 183(4124), 482–488.
pubmed.ncbi.nlm.nih.gov(opens in a new tab)
Baddeley, A. (1994). The magical number seven: still magic after all these years? Psychological Review, 101(2), 353–356.
pubmed.ncbi.nlm.nih.gov(opens in a new tab)
Cowan, N. (2001). The magical number 4 in short-term memory: a reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114.
pubmed.ncbi.nlm.nih.gov(opens in a new tab)

Show all 15 sources

Cowan, N. (2010). The magical mystery four: how is working memory capacity limited, and why? Current Directions in Psychological Science, 19(1), 51–57.
pmc.ncbi.nlm.nih.gov(opens in a new tab)
Ericsson, K. A., Chase, W. G., & Faloon, S. (1980). Acquisition of a memory skill. Science, 208(4448), 1181–1182.
pubmed.ncbi.nlm.nih.gov(opens in a new tab)
Morra, S., Patella, P., & Muscella, L. (2024). Modelling working memory capacity: is the magical number four, seven, or does it depend on what you are counting? Journal of Cognition, 7(1), 60.
journalofcognition.org(opens in a new tab)
Norris, D. G., Hall, J., & Gathercole, S. E. (2019). Can short-term memory be trained? Memory & Cognition, 47(5), 1012–1023.
pmc.ncbi.nlm.nih.gov(opens in a new tab)
Tufte, E. The magical number seven, plus or minus two: not relevant for design — including George Miller's 1998 letter to Mark Halpern.
Archived copy · web.archive.org(opens in a new tab)
Nielsen, J. (2009). Short-term memory and web usability. Nielsen Norman Group.
nngroup.com(opens in a new tab)
Moran, K. (2016). How chunking helps content processing. Nielsen Norman Group.
nngroup.com(opens in a new tab)
Straub, K. (2003). Breadth vs. depth. Human Factors International — summarising Larson, K. & Czerwinski, M. (1998), CHI '98.
humanfactors.com(opens in a new tab)
Myth #23: choices should always be limited to 7±2. UX Myths.
uxmyths.com(opens in a new tab)
Miller's Law. Laws of UX.
lawsofux.com(opens in a new tab)
Miller's law — the four unrelated principles carrying the name. Wikipedia.
en.wikipedia.org(opens in a new tab)