The multi-store model of memory divides memory into three stores: the sensory register, the short-term store and the long-term store. Richard Atkinson and Richard Shiffrin published it in 1968. New information enters the sensory register first and decays there very quickly. A selection of it goes into the short-term store. Without rehearsal, the authors wrote, information in that store is lost within about 15 to 30 seconds. Information is copied from the short-term store into the long-term store, which they called "a fairly permanent repository". The model is also called the modal model or the Atkinson-Shiffrin model.
Textbooks usually show the model as three boxes with arrows and fixed numbers. The 1968 chapter says more than that, and in some places it says something different. The authors saw the control processes as the important part: rehearsal, coding and search. Many of the famous numbers also come from other studies. This article gives the model in Atkinson and Shiffrin's own words and the origin of the numbers. It also gives the strengths and weaknesses that exam questions ask about.
- Atkinson and Shiffrin published the model in 1968 as a book chapter of about 100 printed pages. A 2019 review says that the chapter was mainly about control processes, not only about the three stores.
- In 1968 the authors already called the short-term store the person's "working memory". Baddeley and Hitch published their working memory model six years later.
- The number 7 ± 2 comes from George Miller (1956), not from Atkinson and Shiffrin. Miller himself suspected that the seven was a coincidence.
- The authors described the long-term store as "relatively permanent", not as unlimited. Information in it can change, or it can become temporarily impossible to retrieve.
- The criticism since 1970 is strong, and many textbooks present the model as disproved. But its central ideas are still part of many current theories of memory.
The multi-store model in one table
| Sensory register | Short-term store | Long-term store | |
|---|---|---|---|
| What it holds | An accurate image of the stimulus, for example a visual image | A small selection; the authors studied mainly a verbal store | Information from all the senses |
| How long | For vision, several hundred milliseconds | About 15 to 30 seconds without rehearsal | Relatively permanent |
| How information is lost | Decay, and masking by new stimuli | Decay; in the rehearsal buffer, each new item pushes an old item out | Change, interference, or a retrieval that fails for a time |
| Main control process | Selection of the parts that go on | Rehearsal and coding | Search at retrieval |
The authors' first figure shows the three stores as boxes, one below the other. One arrow goes from the sensory register to the short-term store. Two arrows connect the short-term store and the long-term store, one in each direction. A dashed line goes from the register directly to the long-term store. About this line, the authors wrote: "we do not know whether such transfer occurs".
Where the model came from
The chapter is called "Human memory: A proposed system and its control processes". It appeared in 1968 in volume 2 of the series The Psychology of Learning and Motivation, on pages 89 to 195. Atkinson and Shiffrin worked at Stanford University. The first part of the chapter gives a general framework for memory. The second part tests specific models from this framework in experiments.
The basic idea was not new. Kenneth Malmberg, Jeroen Raaijmakers and Richard Shiffrin name several sources in their 2019 review. The psychologist William James distinguished short-term and long-term memories in 1890. The experiments in our article on Ebbinghaus's forgetting curve also informed the model. In 1958, Donald Broadbent used the terms primary and secondary memory again, and he proposed rehearsal as a way to reactivate information. In 1965, experiments by Waugh and Norman supported the view that new items push old items out of short-term memory. In that view, forgetting comes from the new items, not from the time that passes.
The model later became known as the modal model. According to Alan Baddeley, Graham Hitch and Richard Allen, this term "appears to have been originally proposed by Murdock (1967)". The same authors write that Atkinson and Shiffrin summarised the many advances of the previous decade in one coherent framework. They counted more than 10,000 citations of the chapter on Google Scholar by October 2018.
The three stores, as Atkinson and Shiffrin described them
The sensory register
Every stimulus is first registered in the matching sense. For vision, Atkinson and Shiffrin relied mainly on the experiments that George Sperling published in 1960 and 1963. In the partial-report method, people saw a display for a very short time, usually three rows of four letters and numbers. Then a signal told them which row to report. When the signal came almost at once, the report was accurate. When the signal came later, much of the information was lost.
The authors concluded that a visual stimulus leaves "a more or less photographic trace". This trace decays within several hundred milliseconds, and new stimuli can mask it or replace it. About the other senses, the authors were careful: "There is not much one can say about registers in sensory modalities other than the visual." If a study guide gives a fixed duration for hearing, that number does not come from the 1968 chapter.
The short-term store
The authors described the short-term store on the first pages of the chapter: "The short-term store is the subject's working memory; it receives selected inputs from the sensory register and also from long-term store." In the same passage, information in this store "decays completely and is lost within a period of about 30 seconds". But a control process called rehearsal can keep a limited amount of information there "as long as the subject desires". Two pages later, the authors give "about 15–30 seconds". They add that the exact rate is difficult to estimate, because people rehearse the material by themselves.
The authors also kept two terms apart. The short-term store (STS) is a part of their model. Short-term memory (STM) is the name for experiments that test memory over short periods. "According to our general theory, both STS and LTS are active in both STM and LTM experiments," they wrote. For this reason, this article calls the parts of the model stores.
The authors studied mainly one short-term store for sounds and words. They called it the auditory-verbal-linguistic store, or a-v-l. A word that you read can enter this store as a sound. For its duration, they used an experiment by Peterson and Peterson (1959). People heard three consonants and then counted backward by threes. After 3 to 18 seconds, they had to recall the consonants. Correct recall fell quickly and reached about 8 per cent at 15 to 18 seconds. The authors assumed that the counting task only stopped rehearsal. Under this assumption, a group of three consonants decays from the short-term store within about 15 seconds.
For capacity, the authors started from rehearsal. The number of items that a person can keep by rehearsal depends on the rate of decay. The digit span gives a lower limit: the longest series of digits that a person repeats without errors. According to the authors, this length is "usually in the range of five to eight numbers".
The long-term store
For Atkinson and Shiffrin, the long-term store is "a fairly permanent repository". Information in it is "relatively permanent (although it may be modified or rendered temporarily irretrievable as the result of other incoming information)". In their figure, a note at the long-term store names decay, interference and loss of strength. The common phrase "unlimited and permanent" therefore says more than the authors said.
The model treats retrieval from the long-term store as a search, and cues guide the search. Because the search depends on chance, a memory can be missing at one moment and come back later. In later work, Shiffrin and his colleagues explained forgetting from the long-term store mainly by interference and failed searches, not by decay. Malmberg and colleagues summarise it this way. Our article on proactive and retroactive interference shows how interference works.
Control processes: the part textbooks leave out
Atkinson and Shiffrin divided memory along two dimensions. The first dimension is structure: the fixed stores, which work the same way in every situation. The second dimension is the control processes. The person selects them and uses them as the task requires. The authors used a computer analogy. The stores are like the hardware. The control processes are like the programs that a programmer writes at the console.
According to the 2019 review, this was the core of the chapter. Malmberg, Raaijmakers and Shiffrin write that most citations of the chapter refer to the distinction between short-term and long-term memory. "However, the focus of the chapter's 100 print pages was an investigation of the role of control processes in all memory systems for both storage and retrieval." Shiffrin is a co-author of this review, so here he describes his own work. Three control processes are important for an exam answer.
The rehearsal buffer
Rehearsal is the control process that the authors studied in the most detail. With a long list, a person cannot rehearse all the items. One possible strategy is a buffer with a fixed number of places. The authors compared it to "a bin always containing exactly n items; each new item enters the bin and knocks out an item already there". They called it the rehearsal buffer.
The buffer is not a fixed part of memory. "In our view, the maintenance and use of the buffer is a process entirely under the control of the subject," the authors wrote. When coding or a search of the long-term store helps more, they expected the buffer to become smaller. So the buffer is a strategy inside the short-term store. It is not the short-term store itself.
Transfer means copying, not moving
The arrows in the model do not mean that information moves. The authors used transfer "to mean the copying of selected information from one store into the next without removing this information from the original store". The information stays in the first store and decays there by the rules of that store. Transfer to the long-term store continues for the whole time that the information is in the short-term store.
Information also flows back. In problem solving, in hypothesis testing and in "thinking" in general, the person brings information from the long-term store into the short-term store. Even a word that you read reaches the verbal short-term store only after a search of the long-term store has found its verbal form.
Rehearsal or coding
The authors described two ways to store information. They wrote: "When the subject is concentrating upon rehearsal, the information transferred would be in a relatively weak state and easily subject to interference. On the other hand, the subject may divert his effort from rehearsal to various coding operations which will increase the strength of the stored information."
In their terms, coding is "a select alteration and/or addition to the information in the short-term store as the result of a search of the long-term store". It often uses strong associations that are already in the long-term store. The criticism by Craik and Lockhart later returns to this point.
A multi-store model example: learning a term from your notes
The example below shows how the model describes one act of learning. It illustrates the model. It is not an experiment.
- 1.You read the term "cognitive dissonance" in your notes. The image of the letters enters the sensory register first.
- 2.A search of the long-term store finds the verbal form of the words. Only then does the term enter the verbal short-term store as a sound.
- 3.You repeat the term silently several times. This rehearsal keeps the term in the short-term store while nothing distracts you.
- 4.While the term is in the short-term store, a copy of it goes into the long-term store. If you only rehearse, this trace is relatively weak, according to the authors.
- 5.You connect the term to your own experience, for example a purchase that you rated more highly after you made it. This is coding, and the trace in the long-term store becomes stronger.
- 6.In the exam, you search the long-term store for the term. The exam question is the cue that guides the search.
Our article on cognitive dissonance explains the example in step 5.
Where the famous numbers come from
Study materials on the multi-store model often give fixed numbers for the stores. Many of these numbers come from other studies. The table shows the origin of the best-known ones.
| Statement in study materials | Origin | What Atkinson and Shiffrin wrote |
|---|---|---|
| Short-term memory holds 7 ± 2 items | George Miller, 1956 | Digit span is usually 5 to 8 digits, as a lower limit for rehearsal |
| Short-term memory holds about 4 items | Nelson Cowan, 2001 | This paper appeared 33 years after the chapter |
| Information stays in short-term memory for about 18 seconds | Peterson and Peterson, 1959 | About 15 to 30 seconds without rehearsal |
| The sensory register holds information for less than a second | Sperling's experiments on vision | For vision, several hundred milliseconds; "not much" about the other senses |
The 18 seconds are the longest interval in the Peterson and Peterson experiment. They are not a measured duration of the store.
George Miller began his 1956 paper with the sentence: "My problem is that I have been persecuted by an integer." He then separated two limits: "Absolute judgment is limited by the amount of information. Immediate memory is limited by the number of items." So a person who groups items into larger units, called chunks, can remember more. At the end of the paper, Miller wrote that he suspected the recurring seven was "only a pernicious, Pythagorean coincidence".
In 2001, Nelson Cowan wrote that Miller's seven "was meant more as a rough estimate and a rhetorical device than as a real capacity limit". Other researchers had suggested a more precise limit of only three to five chunks. Cowan concluded that a central capacity limit averages about four chunks. Fergus Craik and Robert Lockhart showed in 1972 how much the estimates vary. Estimates of primary memory gave two to four words, and memory spans gave five to nine items. When the words formed a sentence, young people could repeat up to 20 words accurately.
Strengths: the evidence for separate stores
Atkinson and Shiffrin had to justify two separate stores. Some researchers had argued against a division of memory, for example Melton (1963) and Postman (1964). The authors turned to "what is perhaps the single most convincing demonstration" of the division: the patients that the neuropsychologist Brenda Milner had described. These patients had surgical lesions in the hippocampal region on both sides of the brain.
Milner described the pattern. Immediate registration of new information was normal, for example on a digit span test. The patients kept material that they could rehearse in words for many minutes. When something interrupted the rehearsal, they forgot the material at once. They seemed largely unable to add new information to the long-term store. Atkinson and Shiffrin concluded: "Apparently, a short-term store remains to the patients, but the lesions have produced a breakdown either in the ability to store new information in long-term store or to retrieve new information from it." Our article on retrograde and anterograde amnesia explains the difference between the loss of old memories and a failure to form new ones.
The authors also named the limits of this evidence. The patients were "abnormal to begin with". The operations stopped after the memory defect was discovered, "leaving only a few subjects". And the results varied, for example with the size of the lesion.
A second common argument was the recency effect. In the free recall of a word list, people remember the last words especially well. Baddeley and colleagues note that researchers first saw this effect as a hallmark of the short-term store, and they cite Glanzer and Cunitz (1966).
Weaknesses: the main criticisms
The main objections came quickly and from several directions.
- A patient with a severely impaired short-term store. In 1970, Shallice and Warrington described a patient with a digit span of only two. The patient's long-term store appeared normal. But in the model, information reaches the long-term store through the short-term store. Craik and Lockhart therefore saw the case as evidence "against the idea that information must necessarily 'pass through' STS to enter LTS".
- Short-term tasks are not pure measures. Keppel and Underwood (1962) found that people did better on the first trials of a Peterson session than on later trials. This is a sign of proactive interference. Baddeley and colleagues conclude that the Peterson task "does in fact depend heavily on long-term storage". Atkinson and Shiffrin had already placed this effect in the long-term store. But their 15-second estimate depended on the assumption that the counting task only stopped rehearsal.
- The recency effect is not a reliable sign. Baddeley and colleagues note that it also occurs in long-term memory tasks.
- Levels of processing instead of stores. In 1972, Craik and Lockhart proposed a different framework. For them, the memory trace is "a byproduct of perceptual analysis". The trace lasts longer when the stimulus is analysed more deeply, for example for its meaning rather than for its sound. They saw a "warning sign" in "the progressively greater part played by 'control processes'" in the newer versions of multi-store models.
- A working memory with several parts. Baddeley and Hitch tested whether the short-term store really works as a working memory. People held digit sequences while they did tasks of reasoning, comprehension and learning. Performance dropped, but "the decrements were far less than anticipated". In 1974 they proposed a model with several components: a verbal loop (first called the articulatory loop, later the phonological loop) and a central executive. They mentioned a visuo-spatial system only briefly in 1974 and developed it later. An episodic buffer followed in 2000.
A newer objection goes further. Some researchers, among them Cowan, see short-term storage as simply the active part of long-term memory, not as a separate store. Baddeley and colleagues give several reasons to question this view. The question is open.
Is the multi-store model still accepted?
The answer depends on whom you ask. Malmberg, Raaijmakers and Shiffrin write that "most introductory text books published in the past couple of decades present the dual-store model as disconfirmed in one way or another". But comprehensive theories of memory still "adopt the central tenets of the approach, such as a dual store framework, and the fundamental importance of control processes". They see the debate with Baddeley and Hitch as "one more of perspective than reality".
They also answer Craik and Lockhart. The 1968 chapter itself predicted that coding gives stronger traces than rehearsal alone. In 2013, Lehman and Malmberg reanalysed data from Craik and Tulving. They found that more maintenance rehearsal did improve free recall. This defence comes from Shiffrin and his colleagues themselves.
Baddeley, Hitch and Allen write that the separation of the three stores "has continued to be well supported". But research has divided the stores further. The short-term store became a working memory system with several parts. Long-term memory is now divided, for example, into episodic and semantic memory and into implicit and explicit memory.
Atkinson and Shiffrin did not promise a finished theory. In their words, the first sections of the chapter "do not present a finished theory; instead they set forth a general framework within which specific models can be formulated". As a framework, the multi-store model is still in use. As an exact description of three separate boxes, most textbooks treat it as out of date.
What the model means for learning psychology terms
For learning, the main lesson is the difference between rehearsal and coding. When you only repeat a definition silently, you keep it in the short-term store. But according to Atkinson and Shiffrin, this gives only a relatively weak trace in the long-term store. Craik and Lockhart came to a similar conclusion from a different direction: the more deeply you process a stimulus, the longer the trace lasts. For learning, this means that you process the meaning of a term, not only its sound.
In practice, connect a new term to an example from your own life. Compare it with a similar term, for example the short-term store with the long-term store. Then test yourself instead of reading the definition again. Our article on spaced repetition explains why it helps to spread the practice over several days. Our guide to memorising psychology terminology shows how to use both methods for an exam.
Psychology World is built for this kind of repeated practice. You look up a term and save it to one of your own lists. Then you drill the list as flashcards, by typing the answer, or as multiple-choice questions once the list has four or more terms. The daily word notification also shows you one term each day, without a search. The four steps of the learning loop and what each study mode does are on the home page.
Frequently asked questions
Is the multi-store model the same as the modal model?
Yes. Both names usually mean the 1968 model of Atkinson and Shiffrin, with a sensory register, a short-term store and a long-term store. A third name is the Atkinson-Shiffrin model. In 1972, Craik and Lockhart used the term multistore models more generally, for all models that divide memory into separate stores.
How long does information stay in short-term memory?
According to Atkinson and Shiffrin, verbal information in the short-term store decays within about 15 to 30 seconds without rehearsal. With rehearsal, a person can keep a limited amount for as long as they want. The common figure of 18 seconds is the longest interval in the Peterson and Peterson experiment. Baddeley and colleagues add that this task also depends heavily on the long-term store.
How is the multi-store model different from the working memory model?
The multi-store model has one short-term store, which Atkinson and Shiffrin called the person's working memory in 1968. Baddeley and Hitch found that this store helps reasoning and comprehension less than the modal model suggested. In 1974 they developed it into a working memory with several parts: a phonological loop, a visuo-spatial system and a central executive. An episodic buffer followed in 2000. Malmberg and colleagues see the working memory model as an instance of the modal model's short-term store and control processes.
Can information reach long-term memory without rehearsal?
In the model, yes. Any information in the short-term store goes to the long-term store to some degree during its stay. Malmberg and colleagues describe this transfer as a mixture of automatic transfer and controlled processes such as rehearsal. Whether information can go directly from the sensory register into the long-term store, the authors left open in 1968.
Sources
- 1.Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The Psychology of Learning and Motivation (Vol. 2, pp. 89–195). Academic Press.
- 2.Malmberg, K. J., Raaijmakers, J. G. W., & Shiffrin, R. M. (2019). 50 years of research sparked by Atkinson and Shiffrin (1968). Memory & Cognition, 47(4), 561–574.
- 3.Baddeley, A. D., Hitch, G. J., & Allen, R. J. (2019). From short-term store to multicomponent working memory: The role of the modal model. Memory & Cognition, 47(4), 575–588.
- 4.Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97.
- 5.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.
- 6.Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671–684.