Working Memory vs. Short-Term Memory: The Difference That Matters

"Short-term memory" and "working memory" are often used interchangeably in casual writing, and even in some popular science coverage. They are not, however, the same thing. The distinction matters because the two systems have different properties, develop differently across the lifespan, are impaired by different clinical conditions, and respond differently to training. This article explains the distinction, why it matters for the games on this site, and what the underlying cognitive architecture looks like.

Short-term memory: passive storage

Short-term memory refers to the temporary storage of information — holding something in mind for a few seconds, without manipulating it. The classic example is remembering a phone number you just heard long enough to dial it. The digit-span task (see our Number Memory tool) measures short-term storage capacity: how many digits can you hold in this passive buffer?

George Miller's famous 1956 paper, "The Magical Number Seven, Plus or Minus Two," referred to short-term storage capacity. Miller surveyed a range of tasks — digit span, letter span, binary-digit span, absolute judgment of pitch or loudness — and found that they all clustered around a capacity of about seven items, with most people falling between five and nine. The "plus or minus two" referred to this range. (Modern research has refined this; the true capacity for unchunked information is probably closer to four items, with the higher numbers reflecting unconscious chunking.)

Short-term memory has a limited duration as well as a limited capacity. Without active rehearsal, information decays within 15-30 seconds. The classic demonstration is the Brown-Peterson task: give someone three consonants to remember, then make them count backward by threes from a random number for a varying delay. After 18 seconds of counting, recall is near zero — the rehearsal was blocked, and the contents decayed.

Working memory: storage plus manipulation

Working memory is the system that not only holds information temporarily but also manipulates it — updating it, reordering it, integrating it with new input, suppressing irrelevant items. The term was popularized by Alan Baddeley and Graham Hitch in 1974, who proposed a multi- component model that has dominated the field ever since.

Baddeley's model has four components:

• The phonological loop — stores verbal and acoustic information for short periods. Sub-vocal rehearsal (silently repeating "47, 92, 18…") refreshes this store. This is essentially short-term memory for verbal material.
• The visuospatial sketchpad — stores visual and spatial information. This is what you use to mentally rotate an object, remember a route, or hold a grid of locations in mind (as in our Memory Grid game).
• The episodic buffer — integrates information from the phonological loop, the visuospatial sketchpad, and long-term memory into a coherent episode.
• The central executive — the attentional controller that decides what to attend to, what to update, what to suppress. This is the "manipulation" part of working memory.

The key insight is that working memory is not just a bigger short-term store. It is a system that operates on stored information. Tasks that measure working memory — backward digit span (repeat digits in reverse order), N-Back (track what was N items back as new items arrive), operation span (do mental arithmetic while remembering words) — all require both storage and manipulation. Tasks that measure pure short-term memory — forward digit span, simple Corsi block-tapping — require storage only.

Why the distinction matters

Different capacities

Forward digit span (pure short-term storage) averages about 7 in healthy adults. Backward digit span (working memory, because you must reverse the sequence) averages about 5-6 — a full digit less. The manipulation costs capacity. This is also why N-Back feels so much harder than Number Memory even though both involve remembering sequences: N-Back requires continuous updating, while Number Memory requires only static storage.

Different developmental trajectories

Short-term storage capacity is largely adult-level by early adolescence. Working memory capacity — especially the central executive component — continues to develop into the early twenties, paralleling the development of the prefrontal cortex. This is one reason teenagers can have adult-level digit spans but still struggle with tasks that require sustained attentional control.

Different clinical sensitivities

Working memory, particularly the central executive, is sensitive to a wide range of clinical conditions: ADHD, traumatic brain injury, schizophrenia, depression, Parkinson's disease, and early-stage dementia. Short-term storage is less sensitive to these conditions. This is why clinicians use working-memory tasks (like backward span or N-Back) as part of cognitive assessment batteries — they detect impairments that pure storage tasks would miss.

Different responses to training

Both short-term and working memory can be improved by training on the specific task. The transfer question — does training on a working-memory task improve general cognitive ability? — is contested, as our article on whether brain games work explains. But the practice effects on the trained tasks themselves are real and well-documented.

How the games on this site map onto the distinction

• Pure short-term storage: Number Memory (forward digit span).
• Visuospatial short-term storage: Memory Grid (Corsi block-style, static sequence).
• Working memory (storage + manipulation): Dual N-Back (continuous updating of two streams).
• Working memory + cognitive control: Stroop Test (response inhibition while processing).

The differences in difficulty across these games are not random — they reflect the underlying cognitive demands. If you find N-Back much harder than Number Memory, that's expected: N-Back loads the central executive heavily, while Number Memory loads only the phonological loop.

The larger picture

Working memory is one of the best predictors of general cognitive ability — better than short-term memory, and roughly on par with measures of processing speed. This is one reason it has attracted so much research attention: if working memory could be improved, and if that improvement transferred to broader ability, the practical implications would be large. The current state of evidence suggests the first part is possible (training improves the trained task) but the second part is mostly not (transfer to general ability is weak). We tell you this honestly, as we do elsewhere on the site, because overstating the case would be exactly the kind of behavior that has given the brain-training niche a bad reputation.

Sources

• Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review.
• Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The Psychology of Learning and Motivation.
• Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences.
• Engle, R. W. (2002). Working memory capacity as executive attention. Current Directions in Psychological Science.