What Is Chunking in Memory? How It Works

Chunking is grouping separate pieces of information into larger units that mean something to you, so your short-term memory can hold more. The letters PBSBBCCNN make a hard nine-letter string, but adults remember them much better once they see three TV networks: PBS, BBC and CNN (Feigenson & Halberda, 2008). Short-term memory seems to have room for only a handful of chunks at a time, about seven in George Miller’s 1956 estimate and about four in Nelson Cowan’s newer one (Cowan, 2001). Making each chunk bigger is how you fit more in.

How chunking works

A chunk can be a single letter, digit or word, or a group of them that you’ve learned to treat as one unit (Cowan, 2010). What turns a group into a chunk is knowledge you already have. Chunking makes things easier to memorize because it draws on what you know to cut down how much you have to hold (Chekaf et al., 2016). The researchers give the string IBMCIAFBI as an example: if you know IBM, CIA and FBI, nine letters become three chunks (same source).

Miller described how this builds with practice. Someone just starting to learn radiotelegraph code hears each dit and dah as a separate chunk. Before long the sounds group into letters, then words, then whole phrases (Miller, 1956).

You don’t need to be taught to do it. In the 2008 study, 14-month-old infants could track only three hidden objects at once, but they went past that limit when perceptual, conceptual, linguistic or spatial cues helped them split a larger set into smaller groups (Feigenson & Halberda, 2008).

Miller’s “magical number seven”

Miller’s paper opens with a joke: “I have been persecuted by an integer” (Miller, 1956). For many kinds of material, he wrote, the span of immediate memory “is about seven items in length” (same source).

What mattered was what counts as an item. With binary digits the span was about nine, and with one-syllable English words it fell to about five. That gap is much smaller than you’d expect if memory were limited by the amount of information (Miller, 1956). So Miller concluded that immediate memory is limited by the number of chunks, not bits, and that “we can increase the number of bits of information that it contains simply by building larger and larger chunks” (same source).

He called that process recoding. In his example, 18 binary digits (0s and 1s), regrouped in threes and renamed as octal digits, become 6 digits, well within the span (Miller, 1956). A researcher named Sidney Smith who trained himself on this method could repeat back about 40 binary digits without error (same source). Miller called recoding “an extremely powerful weapon” and said the most common kind is putting things into words, as when we retell a story “in our own words” (same source).

Miller was also careful about the number itself. He warned that the span of immediate memory and the span of absolute judgment, which both come out near seven, are “quite different kinds of limitations,” and he suspected the repeated seven might be “a pernicious, Pythagorean coincidence” (Miller, 1956).

Seven or four? What later research found

The two best-known estimates disagree. In 2001, Cowan wrote that Miller’s seven “was meant more as a rough estimate and a rhetorical device than as a real capacity limit.” He pulled together a wide range of studies to argue for a central limit averaging about four chunks (Cowan, 2001). In 2010 he put it at 3 to 5 meaningful items in young adults (Cowan, 2010).

The newer estimate is the one to use for the underlying limit, but both numbers hold up because they measure different things. Cowan’s central limit shows up only when rehearsal is prevented (Cowan, 2010). Silent rehearsal alone can keep about 2 seconds’ worth of speech in mind (same source). In “running span” tasks, where you don’t know when the list will end, people recall only the last 3 to 5 digits (Cowan, 2010). When the items are longer chunks, such as idioms or short sentences, young adults recall only 3 or 4 of them (same source). Our digit span test guide shows what typical scores look like.

Classic evidence for chunking

Chess masters

In Chase and Simon’s 1973 study, a master, a class A player and a beginner tried to recall chess positions. The master did better only with positions from real games, not with random ones (Küchelmann et al., 2024). Chase and Simon put the master’s edge down to thousands of chunks (familiar clusters of pieces) stored in long-term memory (Gobet & Simon, 1998). When Gobet and Simon repeated the work in 1998, they concluded that chunks are psychologically real and that the two-second pause used to mark where one chunk ends holds up. Their masters used much larger chunks than the master in the 1973 study (same source).

Later work refined one part of the original finding. Looking across studies that showed positions for 3 to 10 seconds, Gobet and Simon found that strong players usually still beat weak players on random positions, though by much less than on real game positions (Gobet & Simon, 1996).

S.F. and 79 digits

In a 1980 study, a participant known as S.F. practiced for more than 230 hours and raised his digit span from 7 to 79 digits (Ericsson, Chase & Faloon, 1980). A long-distance runner, he recoded groups of digits as running times he already knew (Norris et al., 2019). The gain didn’t carry over: with letters instead of digits, his span stayed at seven the whole time (same source).

Everyday examples

• Acronyms: letter strings like PBS, BBC and CNN, or IBM, CIA and FBI, collapse into one chunk each (Feigenson & Halberda, 2008; Chekaf et al., 2016).
• Phone and ID numbers are usually shown in groups of three to four items, and researchers say the grouping helps people hold the numbers in mind and may also help them learn the numbers (Guitard et al., 2022).
• Shopping lists: to remember bread, milk and pepper, Cowan suggests picturing bread floating in peppery milk, which makes three words one image (Cowan, 2010).

How to use chunking

1. Break long strings into groups of three or four. That’s how phone and ID numbers are usually laid out (Guitard et al., 2022).
2. Look for patterns you already know, such as familiar acronyms. S.F. relied on race times he knew as a runner (Norris et al., 2019).
3. Join items into one picture. A single image can hold several words, like the bread in peppery milk (Cowan, 2010). For places in order, picture a path through them (same source). The memory palace technique takes that idea further.
4. Put it in your own words. Miller saw rephrasing as the recoding we do most often (Miller, 1956).
5. Look at each group together. In word-list experiments, grouping helped later recall only when people saw a group’s words at the same time and had more time to process them (Guitard et al., 2022).

You can practice grouping with sequence games such as Simon.

The limits of chunking

• You can only chunk with patterns already in long-term memory (Chekaf et al., 2016). With random chess positions, strong players’ advantage over weak players is much smaller than with positions from real games (Gobet & Simon, 1996).
• The gains are narrow. S.F.’s 79-digit span did nothing for his memory of letters (Norris et al., 2019).
• The core limit stays put. In Cowan’s word-pair experiments with rehearsal blocked, people recalled about 3 units whether those units were single words or learned pairs (Cowan, 2010). Chunking makes each unit carry more, but it doesn’t add slots.
• Some kinds of grouping helped immediate recall but not delayed recall (Guitard et al., 2022).