Mark ở vũ trụ khác · Another Universe Papers That Changed How We Learn #2 · Short-term memory · Chunking

The Magical Number Seven

George A. Miller · "The magical number seven, plus or minus two" · Psychological Review, 1956

The paperPsychological Review 63(2), pp. 81–97, March 1956
Main ideaimmediate memory holds about 7 items, and an "item" is a chunk, not a bit of information
Revised byCowan (2001): when chunking is blocked, only about 4
LabelCaution: chunking is solid, the number seven is not

Source: Miller (1956), Psychological Review 63(2), 81–97; Cowan (2001), Behavioral and Brain Sciences 24(1), 87–114. The label is Mark's assessment.

Re-read in 1 minute

Essence

About 6 levelsHear a tone and name its pitch: people start confusing them beyond about 6 levels.
About 7 items backHear a list and repeat it: young adults hold about 7 digits.
Counted in chunksThe limit is in chunks, not in amount of information. Group things and you hold more.
Really closer to 4Cowan (2001): with rehearsal and chunking blocked, the limit is about 3 to 5 chunks.
Not a "7-item rule"The paper never says menus or slides need 7 items at most. Miller even called the sevens a coincidence.

Source: Miller (1956); Cowan (2001), abstract on PubMed 11515286; Tufte (2003) on the "7-item rule" (secondary source).

Part 1

The paper's question

How much can a head hold at once? Miller saw the number seven turning up everywhere, then pointed to something more important than the number: we count in chunks, and whoever knows how to chunk remembers more.

The question is Mark's summary of the paper's opening and closing.

Part 1Absolute judgmentNaming a tone or a brightness: about 2.5 bits, roughly 6 levels.
→
Part 2Immediate memory spanRepeating a list: about 7 items, nearly regardless of how much each item carries.
→
Part 3ChunkingRecode information into bigger chunks to get past the limit.
Part 2

The story of the discovery

An American psychologist, the heyday of information theory, and a number that would not let go.

My problem is that I have been persecuted by an integer.Opening line of the paper, Miller (1956). He writes that the number had followed him for seven years, intruding into his own data and the pages of leading journals.
1948 onward · Information theoryThe idea of measuring information in bits and of "channel capacity" starts to enter psychology. Miller uses this frame to look at a person as a communication channel.
15/4/1955 · PhiladelphiaMiller gives an invited address to the Eastern Psychological Association.
3/1956 · Psychological ReviewThe talk is printed as a paper, volume 63, pages 81–97.
11/9/1956 · A symposium at MITMiller later wrote (2003) that he dated the birth of cognitive science to this day, the second day of a symposium on information theory.
1991 · A medalMiller receives the US National Medal of Science for his work on how the mind operates.
2012 · DeathMiller dies on 22/7/2012, aged 92, in New Jersey.

Source: the footnote printed in Miller (1956) on the 15/4/1955 address (checked via excerpts of the PDF); Miller (2003), Trends in Cognitive Sciences, via PhilPapers and Lateral magazine (secondary sources); Princeton University news, 26/7/2012 (1991 medal, date of death).

Absolute judgment experiment. A range of tones from low to high. With a few levels, listeners name them correctly. Beyond about 6 levels, they start to confuse them. Transmitted information is about 2.5 bits. hear a tone, name its level (Pollack's experiment, as summarised by Miller) 123456 7+? lowhigh about 2.5 bits ≈ 6 distinguishable levels add more and confusion starts illustration; the number of levels is Miller's summary, not one person's result
A person as a narrow channel. For pitch, discrimination is about 2.5 bits. Averaged over several senses, Miller summarised about 2.6 bits.Source: Miller (1956), section on absolute judgment, summarising Pollack's experiment with tones of 100–8000 Hz; bit values checked via summaries (andymatuschak.org, uxdesign.cc; secondary sources).

Two different limits

Absolute judgment is measured in bits. Memory span is measured in chunks. Both come out near 7, but they are not the same thing.

A witty talk

Miller writes as if telling the story of being "persecuted" by a number, and ends by doubting those very sevens.

What got forgotten

The main idea is chunking. What people remember is the number seven.

Source: Miller (1956); "two limits of different kinds" per the paper itself and Wikipedia (secondary source). The last card is Mark's view.

Part 1 of 3

The discovery

1955information theory6 levelsquiz
Part 2 of 3

What it says, how it was received

chunking40 binary digitsmyths
Part 3 of 3

New research, lessons, Mark's score

the number 4slots or resourceshow to chunk
Part 3

What the paper says

In 3 sentences, then in pictures.

  1. When naming a stimulus along one dimension (pitch, loudness...), we can tell apart only about 6 levels.
  2. When hearing a list and repeating it, we hold about 7 items, and this limit is counted in chunks, not in amount of information.
  3. So grouping information into bigger chunks (recoding) is the strongest way past the limit.

The three sentences are Mark's summary of Miller (1956).

Chunking. The twelve digits 1 9 4 5 1 9 7 5 2 0 2 6 are twelve items, beyond the span. Read as the three years 1945, 1975, 2026 they become three chunks, easy to hold. the same 12 digits, two ways to hold them 194519752026 12 items 1945 · independence1975 · reunification2026 · this year 3 chunks Mark's illustration (Vietnamese history years); meaning already in your head turns 4 digits into 1 chunk
Not remembering more, but counting differently. The number of items we can hold is nearly fixed; what can change is how big each item is.Source: chunking and recoding per Miller (1956). The years are Mark's illustration.
Recoding binary digits. Nine digits 1 0 1 0 0 0 1 1 1 split into three groups of three, 101, 000, 111, read as 5, 0, 7. Nine items become three. Sidney Smith trained this way and, as Miller tells it, could recall a string of about 40 binary digits. Sidney Smith's trick: rename binary digits in groups 101 000 111 5 0 7 9 digits → 3 chunks Smith trained himself and recalled a string of about 40 binary digits (as Miller tells it) "101 000 111" is Mark's illustration; the figure 40 has not been checked against the original
Learn the recoding, then make it automatic. At first the recoding is slow and memory barely improves. Once it becomes a reflex, the span jumps.Source: Miller (1956), section on recoding, checked via a discussion list post by Stevan Harnad, University of Southampton (secondary source).
More: why Miller doubted the number seven himself

The paper's ending runs through the sevens: seven wonders, seven seas, seven deadly sins, seven notes, seven days. Then Miller writes that he suspects it is only "a pernicious, Pythagorean coincidence".

In other words, the author himself did not claim seven was the brain's capacity. Cowan (2001) reads it the same way: the seven was a rough estimate and a rhetorical device.

Source: Miller (1956), closing section, quoted via secondary pages (uxdesign.cc and others); Cowan (2001), abstract.

Part 4

How this paper has been judged

Where it appeared, citations, replications, criticism. Each line has a source.

Where, whenPsychological Review, volume 63, issue 2, pages 81–97, March 1956. Originally an address given on 15/4/1955.
CitationsNot retrieved. Google Scholar, Semantic Scholar, OpenAlex were blocked from the writing environment (checked 15/10/2026). Wikipedia calls it one of the most highly cited papers in psychology (secondary source). No number is given.
Replicated?Partly. A digit span of about 7 is still measured. But with rehearsal and chunking blocked, the limit drops to 3–5 chunks (Cowan 2001); lists with no structure hold only about 3 items (Chekaf, Cowan and Mathy 2016).
Main criticismCowan (2001, 2015): seven is rhetorical, the real limit is near 4. Baddeley (1994) asked "still magic?" in Psychological Review itself. Ma, Husain and Bays (2014): there may be no fixed "slots" at all, but a flexibly shared resource.
StandingOne of the best-known papers in psychology (Cowan 2015). Miller received the US National Medal of Science in 1991 for his work on the mind.
VerdictCAUTION Chunking and the existence of a short-term memory limit: solid. Seven as "brain capacity": not solid, closer to 4.

Source: Miller (1956); Cowan (2001), Behavioral and Brain Sciences 24(1); Chekaf, Cowan and Mathy (2016), Cognition; Baddeley (1994), Psychological Review 101(2), per the volume 101 contents (secondary source); Ma, Husain and Bays (2014), Nature Neuroscience 17(3); Cowan (2015), Psychological Review 122(3); Princeton news 26/7/2012. The label is Mark's assessment.

Part 5

Fact-check: what holds up and what doesn't

14 things often said about "the number seven". Tap a row for the reason.

SOLIDPublished in Psychological Review in 1956, from a 1955 talk

The paper's footnote: an invited address to the Eastern Psychological Association, Philadelphia, 15/4/1955.

SOLIDWe can tell apart only about 6 levels of pitch

Miller summarises Pollack's experiment: about 2.5 bits, roughly 6 levels, before confusion begins.

SOLIDYoung adults can repeat about 7 digits

Miller (1956) on immediate memory span. This is a digit-span figure, measured when listeners are free to rehearse and chunk.

SOLIDChunking lets you hold more

Miller (1956); Chase and Simon (1973): strong chess players recall real positions far better, but their edge shrinks a lot for random positions; Chekaf et al. (2016): span grows when a list has structure.

SOLIDWithout chunking, we hold about 3 to 5 chunks

Cowan (2001) brings many kinds of data together and concludes the central limit averages about 4 chunks.

SOLIDVisual short-term memory holds about 4 objects

Luck and Vogel (Nature, 1997): about 4 colours or orientations; each object can carry several features.

CAUTIONSmith recalled about 40 binary digits by recoding

One self-trained person, known only through Miller's account; the figure 40 has not been checked against the original (via Harnad's discussion list, secondary source).

CAUTIONSpan differs by material: digits more than letters, letters more than words

Summaries give about 7 for digits, 6 for letters, 5 for words (secondary sources). The direction is plausible; the exact numbers are unchecked.

CAUTIONWorking memory has a fixed number of "slots"

Ma, Husain and Bays (2014) argue it may be a resource shared flexibly among all items, with quality dropping as items increase. The "slots or resources" debate continues.

CAUTIONDigit span has stayed the same for decades

Gignac (2015) found no trend across 10 normative samples, 1923–2008; Woodley of Menie and Fernandes (2015) re-analysed and found one; Wongupparaj et al. (2017): forward span rose slightly, backward span fell slightly.

FALSE"Miller proved that seven is the brain's capacity"

Miller himself called the sevens "a pernicious, Pythagorean coincidence" (quoted via secondary sources). Cowan (2001): the seven was a rough estimate and a rhetorical device.

FALSE"Short-term memory holds exactly 7 things, plus or minus 2"

With rehearsal and chunking blocked, the limit is about 4 (Cowan 2001). "7 plus or minus 2" describes digit span under easy conditions, not a number of slots in the head.

FALSE"Phone numbers have 7 digits because of Miller's paper"

New York used 2 letters plus 5 digits (7 in all) from 1930; the 1947 North American numbering plan standardised that format, before the paper (Wikipedia, secondary source). No record shows the phone company chose it because of Miller.

FALSE"Menus and slides should have at most 7 items, per Miller's law"

Tufte (2003): the paper neither states nor implies any rule for how much to show in a presentation. What is in front of your eyes need not be remembered; Miller's limit is about memory.

Source: as given in each row; full list in Part 10.

Part 6

What the latest research says

From "7" to "4", then to whether there are slots at all.

Two numbers compared. Miller 1956: about 7 items, with free rehearsal and chunking. Cowan 2001: about 4 chunks, range 3 to 5, with rehearsal and chunking blocked. Miller 1956 ≈ 7 items, rehearsing, chunking Cowan 2001 ≈ 4 chunks (3–5), both blocked the two numbers measure different conditions; each is right for its own
Not a contradiction, two different questions. "How much can you repeat back" differs from "how many places are there in your head".Source: Miller (1956); Cowan (2001), abstract on PubMed 11515286.
1994 · "Still magic?"Baddeley revisits the seven in Psychological Review itself.
1997 · Vision is about 4 tooLuck and Vogel (Nature): about 4 colours or orientations in visual short-term memory.
2001 · The number 4Cowan (Behavioral and Brain Sciences): a central limit of 3–5 chunks, averaging about 4.
2014 · Slots or resources?Ma, Husain and Bays (Nature Neuroscience): perhaps no fixed slots, but a flexibly shared resource.
2015 · Miller in retrospectCowan (Psychological Review): the 1956 paper was followed by about 40 years with little research on the numerical limit.
2016 · Structure raises spanChekaf, Cowan and Mathy (Cognition): lists with no structure hold about 3 items; the more structure to chunk, the bigger the span.
2017 · Across decadesWongupparaj et al. pooled data from 1973–2016: forward span rose slightly, backward span fell slightly.
2018 · A set of benchmarksOberauer et al. (Psychological Bulletin) list the robust short-term memory findings any theory must explain.

Source: Baddeley (1994), Psychological Review 101(2); Luck and Vogel (1997), Nature 390, 279–281; Cowan (2001); Ma, Husain and Bays (2014), Nature Neuroscience 17(3), 347–356; Cowan (2015), Psychological Review 122(3), 536–541; Chekaf, Cowan and Mathy (2016), Cognition; Wongupparaj et al. (2017), Intelligence, as summarised in the data paper on PMC7013124; Oberauer et al. (2018), Psychological Bulletin 144(9), 885–958.

Part 7

Assessment

Scores are Mark's assessment based on the sources in Part 10, not data from a study.

Chunking still holds
5/5
The number seven holds
2/5
Influence on the field
5/5
Understood correctly online
1/5
Strengths
  • Pulled scattered data into one big question
  • Introduced the "chunk", still used today
  • Doubted its own number
Weaknesses
  • The seven was catchy enough to bury the main idea
  • Measured with free rehearsal and chunking
  • Turned into a "design law" it never stated

Miller was right about what matters: we count in chunks. The seven is just a pretty label, and he did not believe it himself.

More: why the "rule of 7" won't die

A number, a scientist's name, and advice you can use at once. Those three put "Miller's law" into design books, presentation courses and skills workshops.

The accurate version is less tidy: the limit is near 4 without tricks, and much larger once you chunk. What is worth teaching is how to chunk, not the number.

"More" is Mark's commentary, based on the sources in Parts 4 and 5.

Part 8

Apply it: break information into small chunks

Doable from tomorrow: vocabulary, numbers, note-taking, exam revision.

Four habits: group in threes and fours; attach to existing meaning; name each chunk; write it down instead of holding it. 1 · groups of 3–40912 345 678not 0912345678 2 · attach meaningdigits into datesmeaning merges items 3 · name the chunkone name, one big idearecall the name, pull the chunk 4 · write it downfree up spacehead to think, paper to hold Don't try to hold more items. Make each item bigger.
Each habit rests on a sourced finding. A limit near 4 chunks (Cowan 2001), chunking extends span (Miller 1956; Chekaf 2016), experts remember in meaningful chunks (Chase and Simon 1973).Source: as in Parts 3 and 6. The phone number is illustrative; grouping in "3–4" is Mark's suggestion based on Cowan's 4.
Groups of 3–4Twelve new words? Split them into 3 themed groups of 4 and learn one group at a time.
Hook it to what you knowTie numbers to years, birthdays, familiar prices. Existing meaning turns many items into one.
Name your chunksTake notes as 3–4 big ideas, each with a short name. Remember the names first, details second.
Write it, don't carry itOn a hard problem, put the steps on paper. Short-term memory is small; don't make it hold and think at once.
Self-check: your memory span, 5 steps

1. Have someone read you 8 random digits, one per second. Repeat them. How many did you get?

2. Try a new list, but this time group them in 3s and 4s in your head. How many now?

3. Try again while saying "la la la" as you listen, so you can't rehearse. How far does it drop?

4. Write today's lesson as exactly 4 named big ideas. Tomorrow, try to recall the 4 names.

5. Compare steps 1 and 2: the gap is the power of chunking.

Step 3 mimics how researchers block rehearsal when measuring the "about 4" limit (Cowan 2001). This is a self-test, not a standard measurement.

Part 9

Read next

Miller (1956), "The magical number seven, plus or minus two"Psychological Review. Written like a talk, witty, readable in an evening.
Cowan (2001), "The magical number 4 in short-term memory"Behavioral and Brain Sciences, with open peer commentary. Long, but the key paper for why 4.
Cowan (2015), "George Miller's magical number of immediate memory in retrospect"Psychological Review. Short, the afterlife of the 1956 paper.
Papers That Changed How We Learn #01: The Forgetting CurveEbbinghaus could recite a 7-syllable list after one reading, more than 70 years before Miller.
Papers That Changed How We Learn #03: Patient H.M. (1957)An operation left a man unable to form new memories, yet still able to learn skills.
Part 10

Sources

The paper itself and peer-reviewed studies come first. Secondary sources are labelled. The writing environment blocks the original-paper pages and citation databases, so the originals were checked through excerpts of those same pages in search results, not opened in full.

  1. 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. doi.org
  2. 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
  3. Cowan, N. (2015). George Miller's magical number of immediate memory in retrospect. Psychological Review 122(3), 536–541.
  4. Baddeley, A. (1994). The magical number seven: Still magic after all these years? Psychological Review 101(2).
  5. Luck, S. J. and Vogel, E. K. (1997). The capacity of visual working memory for features and conjunctions. Nature 390, 279–281. doi.org
  6. Ma, W. J., Husain, M. and Bays, P. M. (2014). Changing concepts of working memory. Nature Neuroscience 17(3), 347–356. doi.org
  7. Chekaf, M., Cowan, N. and Mathy, F. (2016). A study of chunk formation and data compression in immediate memory. Cognition.
  8. Oberauer, K. et al. (2018). Benchmarks for models of short-term and working memory. Psychological Bulletin 144(9), 885–958. doi.org
  9. Chase, W. G. and Simon, H. A. (1973). Perception in chess. Cognitive Psychology 4, 55–81.
  10. Gignac, G. E. (2015), Intelligence; Woodley of Menie, M. A. and Fernandes, H. B. F. (2015); Wongupparaj, P. et al. (2017), Intelligence (as summarised on PMC7013124).
  11. Miller, G. A. (2003). The cognitive revolution: a historical perspective. Trends in Cognitive Sciences.
  12. Secondary sources: Princeton University news, 26/7/2012; Edward Tufte, "The magical number seven, plus or minus two: Not relevant for design" (20/4/2003); Wikipedia (The Magical Number Seven, Plus or Minus Two; Telephone exchange names; All-number calling); Stevan Harnad's discussion list (Southampton); uxdesign.cc; andymatuschak.org; PhilPapers; Lateral magazine.

Citation count: not retrieved (Google Scholar, Semantic Scholar, OpenAlex blocked, checked 15/10/2026).