Atkinson–Shiffrin memory model¶

A three-store human-memory model with task-dependent control processes for transfer and retrieval.

Version: v1 · 2026-09-28 · History
Domain-specific #: 8056
Domain group: Social Sciences
Origin domain: Psychology & Behavioral Sciences
Subdomains: Cognitive Psychology, Memory Research → Psychology & Behavioral Sciences

Core Idea¶

Atkinson and Shiffrin's 1968 memory model proposes three structural components: a brief sensory register, a short-term store used as a temporary workspace, and a long-term store supporting later retrieval. It also distinguishes those structures from flexible control processes such as attention, rehearsal, coding, and search. Information does not simply march through three fixed boxes; selected inputs and strategies influence what is kept, learned, or retrieved.

The model was offered as a framework from which task-specific mathematical accounts and experiments could be developed. Paired-associate work illustrates how rehearsal-buffer assumptions can predict patterns of recall. The model's historical influence does not make its stores literal brain compartments, nor does it settle all later disputes about working memory and encoding routes.

How would you explain it like I'm…

Blink, Notepad, Storage

Two scientists drew a picture of how memory might work. There is a super-quick glimpse part that holds what you see and hear for a blink, a little notepad part for what you're thinking about right now, and a big storage part for things you can remember later. What you pay attention to and repeat to yourself helps decide what gets kept. It's a helpful picture, not real boxes inside your head.

Three Stores of Memory

The Atkinson–Shiffrin memory model is an idea from 1968 about how memory is organized. It has three parts: a sensory register that holds what your senses pick up for a very short time, a short-term store where you work with information for a little while, and a long-term store that keeps things so you can remember them later. It also includes things you do with memory, like paying attention, repeating information, and searching for what you know. Information doesn't just slide through three boxes in a line; what you notice and how you practice changes what you learn. The model is a way to organize ideas and experiments, not a map of real parts of the brain.

Stores-and-Control Memory Model

The Atkinson–Shiffrin model, proposed in 1968, describes memory with three structural parts: a brief sensory register, a short-term store that acts as a temporary workspace, and a long-term store that supports later retrieval. It separates these fixed structures from flexible control processes such as attention, rehearsal, coding and search. So information does not simply march through three boxes in order; which inputs are selected and which strategies are used shape what gets kept, learned or recalled. The model was a framework for building specific mathematical accounts and experiments, such as using a rehearsal buffer to predict recall in paired-associate tasks. Its influence does not mean the stores are literal brain compartments or that it settled later debates about working memory.

The Atkinson–Shiffrin model, proposed in 1968, divides memory into structural components and control processes. The structures are a brief sensory register, a short-term store serving as a temporary workspace, and a long-term store that supports later retrieval. Control processes, including attention, rehearsal, coding and search, are flexible, strategy-dependent operations that determine which inputs are selected, maintained, transferred or retrieved. The model therefore does not claim that information passively marches through three fixed boxes; what reaches long-term memory depends on selection and control. It was offered as a general framework from which task-specific mathematical accounts and experiments could be derived. Paired-associate learning studies illustrate this, with rehearsal-buffer assumptions generating quantitative predictions about recall patterns. Two limits matter: the stores are functional constructs rather than literal brain compartments, and the model's historical prominence does not resolve later disputes about working memory and encoding routes.

Structural Signature¶

Sig role-phrases:

Sensory register — Briefly holds modality-specific incoming information before selected transfer or decay. It is constitutive. Counterfactual: Without sensory registration the model loses its first proposed structural component.
Short-term store — Temporarily holds selected items and serves as a workspace for control processes. It is constitutive. Counterfactual: A two-store account without this central workspace is not the three-store proposal.
Long-term store — Provides relatively durable stored information that can be learned and retrieved. It is constitutive. Counterfactual: Without long-term storage the account cannot explain retention beyond temporary workspace.
Control processes — Task- and person-dependent attention, rehearsal, coding, and retrieval manage traffic among stores. It is constitutive. Counterfactual: Treating all transfers as fixed automatic pipes erases a major distinction in the authors' formulation.
Information-flow links — Specify selected input, learning, and retrieval directions among the proposed stores. It is central. Counterfactual: Three unconnected box labels would not constitute the model's process account.
Empirical task comparison — Connects parameterized versions of the framework to observable learning or recall. It is central. Counterfactual: Without task and observation the framework is a diagram, not an evaluated cognitive account.

What It Is Not¶

Not three brain locations. The stores are theoretical functional components.
Not automatic one-way flow. Strategies and retrieval affect traffic.
Not identical to modern working-memory theories. Later models divide the temporary workspace differently.
Not proof that repetition alone creates all long-term memory. Control processes are varied.
Closest near-miss. The stores are theoretical constructs, not proven one-to-one brain regions; later working-memory models and criticisms can refine or replace specific assumptions.

Scope of Application¶

History of cognitive psychology. Locate the 1968 multi-store framework in memory theory.
Experimental memory research. Derive task-specific predictions from stores and control strategies.
Education about models. Distinguish explanatory construct from directly observed anatomy.
Theory comparison. Contrast three-store assumptions with later working-memory or alternative accounts.

Clarity¶

The distinctive move is two-dimensional: proposed stores are relatively fixed, while control processes vary with the task and learner. Sensory input, temporary availability, and durable retention are separate in the diagram. A predicted recall curve tests a particular implementation, not the existence of literal memory boxes.

Manages Complexity¶

A single recall score mixes perception, attention, temporary holding, encoding, and retrieval. Separating stores from strategies lets a researcher ask which part of an experiment changes, but too many unobserved components can make fit ambiguous. The model is useful as a disciplined hypothesis generator only when each tested prediction and its alternatives are stated.

Abstract Reasoning¶

Specify the sensory input and test schedule.
Identify which proposed store holds information at each step.
State the subject's attention and rehearsal strategy separately from structure.
Model transfer and retrieval links for the task.
Predict observable recall patterns.
Compare data and alternative accounts without reifying the boxes.

Knowledge Transfer¶

The framework applies literally to theoretical accounts of human memory tasks where sensory registration, temporary workspace, durable store, and control processes are the hypothesized mechanisms. A computer cache may resemble a multi-store diagram but does not establish the same psychological theory. The portable modeling lesson is to separate relatively stable architecture from variable strategies; the named model stays historical and cognitive.

Examples¶

Canonical¶

Atkinson and Shiffrin's 1968 chapter models continuous paired-associate study/test trials. A presented cue-response pair enters through registration; selected information occupies a short-term rehearsal buffer; control processes choose how to rehearse; learned pair information is represented in long-term store and retrieved on later test trials. This is an example of how a model is fitted to a task, not direct observation of three physical compartments.

Mapped back: Sensory register → visual presentation of study/test items; Short-term store → temporary rehearsal buffer for recently encountered pairs; Long-term store → durable pair associations available later; Control processes → rehearsal and response strategies; Information-flow links → registration, selected transfer, learning, later retrieval; Empirical task comparison → continuous paired-associate trials used to test a specific model.

Applied / In Practice¶

In a contemporary primary study of paired associates, overt and covert study procedures produced different forgetting-curve shapes. A mathematical account using a short-term rehearsal buffer and long-term memory state predicted those data. The experiment tests a specific process implementation; its fit does not validate every claim about sensory registration or prove that rehearsal is the sole long-term encoding route.

Mapped back: Sensory register → presented word pairs must be perceived, but the study does not isolate this store; Short-term store → rehearsal buffer in the fitted model; Long-term store → long-term memory state in the fitted model; Control processes → overt versus covert rehearsal procedure; Information-flow links → buffer use and longer-term retention; Empirical task comparison → observed distinct forgetting curves under two study conditions.

Structural Tensions¶

T1 — Fixed Architecture versus Variable Strategy. The same three-store sketch can generate different performance through different control processes; box labels alone underdetermine behavior.

Diagnostic: Which results follow from structure and which from chosen rehearsal or coding?

T2 — Temporary Maintenance versus Durable Retention. Rehearsal-buffer occupancy can aid learning but short-term activation is not identical to a durable trace.

Diagnostic: What evidence separates immediate availability from later recall?

T3 — Theoretical Fit versus Literal Anatomy. A successful task model is evidence for an explanatory structure, not a photograph of three discrete neural bins.

Diagnostic: Which claims are tested by this experiment and which remain modeling assumptions?

Structural–Framed Character¶

The Atkinson–Shiffrin model is mixed-framed: its stores and flows can be diagrammed formally, but the proposed architecture is a historical theory about human memory. Evaluative weight: the model's explanatory usefulness is a research judgment, not proof that every mind literally contains three boxes. Human-practice-bound: people remember without theorists, whereas the sensory/short-/long-term divisions and control-process interpretation are model-building choices. Institutional origin: the 1968 research proposal and later experimental tradition fix the named account; an institutional endorsement would not make its mechanisms true. Vocabulary travels: stage, store, and control process can describe computers, but a cache diagram is not evidence for human sensory registration or rehearsal. Import versus recognize: using the same model to organize another memory task is literal application, while labeling a machine's storage tiers “Atkinson–Shiffrin” is analogy.

The portable skeleton is the live parent prime Representation: a structured medium selects features of a target for reasoning. Here a three-store diagram and associated mathematics represent proposed human-memory functions, with control processes affecting task predictions. The historically specific architecture and psychological evidence are the domain accent. Its character: a testable cognitive representation, not a universal memory law or a literal description of every retrieval event.

Structural Core vs. Domain Accent¶

Skeletal core. Staged stores and adjustable control processes jointly generate task predictions. Domain-bound accent. The stores and tasks refer to human sensory registration, short-term retention, and long-term recall in the authors' 1968 proposal. Substitute database cache tiers and the diagrammatic analogy survives, but the psychology model does not. Why not a prime. Its empirical target and historical theoretical commitments are constitutive.

This entry is a kind of Representation.

Current DAG placement. Live prime Representation maps a target to a medium under a stated convention while preserving selected structure. This model maps hypothesized human-memory functions and information flow into a diagrammatic/mathematical system, retaining selected timing and transfer features for reasoning while omitting physiology. That exact child-to-parent relation supports strict subsumption.
Neighboring theories. Later working-memory architectures are revisions or alternatives, not aliases.

Relationships to Other Abstractions¶

Current abstraction Atkinson–Shiffrin memory model Domain-specific

Parents (1) — more general patterns this builds on

Atkinson–Shiffrin memory model is a kind of Representation Prime

The three-store memory theory represents selected human-memory functions and information flow in a diagrammatic and mathematical medium.

Hierarchy path (1) — routes to 1 parentless root

Atkinson–Shiffrin memory model → Representation → Abstraction

Neighborhood in Abstraction Space¶

Atkinson–Shiffrin memory model sits in a crowded region of the domain-specific corpus (39^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Memory Storage, Retrieval & Encoding (9 abstractions)

Nearest neighbors

Exosomatic Memory — 0.90
Storage (memory) — 0.89
Cue-dependent forgetting — 0.88
Time-sharing — 0.86
MUSHRA — 0.86

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

A three-box picture alone. Tell: Omits variable control processes and empirical task commitments.
Brain anatomy map. Tell: Treats functional stores as localized structures without evidence.
Modern working-memory model. Tell: May distinguish executive and modality-specific subsystems absent from the original sketch.
Any rehearsal effect. Tell: Does not by itself validate the whole 1968 architecture.

References¶

Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Atkinson%E2%80%93Shiffrin_memory_model (revision 1369240345).
R. C. Atkinson and R. M. Shiffrin, "Human Memory: A Proposed System and Its Control Processes" (1968), original three-store/control-process account and continuous paired-associate models: https://escholarship.org/uc/item/5kd4s4j3
"Recall of paired-associates as a function of overt and covert rehearsal procedures," Journal of Verbal Learning and Verbal Behavior 7 (1968): 730–736, original experiment and model: https://doi.org/10.1016/S0022-5371(68)80134-3

The original chapter supports a historical model and its testable implementations, not a claim that each store is a discrete anatomical compartment. The separate rehearsal experiment tests one buffer-plus-long-term-state account rather than every premise of the framework.