The working memory test is the cognitive psychology measurement that has done the most work in the last fifty years. Working memory is the small, active mental workspace you use to hold things in mind while doing something else with them: the running total in mental arithmetic, the previous point in a conversation while listening to the next one, the cooking step you're following while watching for the timer.
It is not the same thing as short-term memory, and it's not the same as memory in the colloquial sense. This page is the pillar-level explainer of what working memory tests actually measure, which ones to use for what, and what high or low scores genuinely predict.
Working memory vs. short-term memory
These get used interchangeably in casual writing. They are not the same thing.
Short-term memory is the passive, brief storage of recently-presented information. The classic short-term-memory task is forward digit span — hear digits, repeat them. No transformation, just storage.
Working memory is short-term storage plus active manipulation. The classic working-memory task is backward digit span — hear digits, hold them, reverse them in mind, report them. The reversal is the manipulation. The storage alone wouldn't be enough.
The functional distinction matters because the two have different cognitive correlates. Short-term memory correlates with language and reading scores. Working memory correlates more strongly with fluid reasoning, problem-solving, and academic outcomes.
The modern theoretical framework is Alan Baddeley and Graham Hitch's 1974 multi-component model, introduced in their Psychology of Learning and Motivation chapter. Their model has three (later four) components:
- Central executive — the attentional control system that allocates resources.
- Phonological loop — verbal short-term store with rehearsal.
- Visuospatial sketchpad — visual short-term store.
- Episodic buffer (added 2000) — binds information across modalities into integrated chunks.
Most working memory tests stress one or two of these components. Backward digit span and reading span stress the central executive plus the phonological loop. The Corsi block test stresses the visuospatial sketchpad. The N-back task arguably stresses the central executive most directly.
Which working memory test should you use?
It depends on what you're trying to measure.
For research-grade, validated working memory measurement, the recommended tests come from the complex span family:
- Operation Span (OSPAN) — solve a simple math problem, remember a word, repeat across a series, then report the words in order.
- Reading Span (RSPAN) — judge whether a sentence makes sense, remember the last word, repeat, then report.
- Symmetry Span (SSPAN) — judge a visual pattern's symmetry, remember a spatial location, repeat, then report.
These tests are the modern standard. The Conway, Kane et al. 2005 Psychonomic Bulletin & Review paper is the methodological-user's-guide reference for them. They produce the highest correlations with fluid intelligence and other higher-order cognitive measures.
For fast, scalable, fMRI-friendly working memory measurement, the N-back task is the standard. See the N-back test page for the full explainer. N-back is easier to instrument in scanners and across large samples but correlates somewhat less strongly with other working-memory measures than complex span does.
For clinical working memory measurement, backward digit span is the most-used subtest, partly because it's been in the Wechsler intelligence batteries for decades and has the most published norms.
For rough self-assessment of working memory, any of the above will give you a number. None of them is a stable measure of your "true" working memory capacity from a single sitting — sleep, fatigue, attention, and strategy all move the score by meaningful amounts session to session.
What working memory tests actually predict
Working memory capacity is one of the most-studied cognitive constructs because it correlates with a lot of things.
In large samples, working memory measures correlate:
- Modestly to moderately with fluid intelligence (Gf, Raven's matrices). Kane and Engle's 2002 Psychonomic Bulletin & Review paper argued the relationship runs primarily through executive attention.
- Moderately with reading comprehension. Working memory bottlenecks the integration of information across sentences in complex text.
- Moderately with math achievement. Mental arithmetic and multi-step problem solving load working memory directly.
- Weakly with crystallized knowledge and vocabulary. These rely more on long-term memory and accumulated experience than active working memory.
What it does not strongly predict:
- Specific creative output. Working memory and creativity correlate weakly at best.
- Social skills. Largely separate.
- Specific job performance, except in roles where working memory is directly loaded (e.g., complex multi-step procedures, simultaneous interpretation).
Engle's research program (see his 2002 Current Directions paper) frames working memory capacity as essentially executive attention applied to memory — the ability to keep relevant information active and suppress interference from irrelevant information. That framing has been productive and influential, and it's the one most modern working-memory research operates under.
What "good" looks like
For untrained adults across the main working memory measures:
- Complex span (OSPAN / RSPAN): typical adult scores fall between 20 and 60 (out of about 75), with substantial individual variation.
- N-back: most adults plateau between N=3 and N=4 on single n-back.
- Backward digit span: typically 3–7 digits.
- Corsi block span: typically 4–7 positions.
The honest framing across all these measures: working memory capacity in untrained adults is about 3 to 5 chunks of independent information, with substantial individual variation around that average. This number — small by intuition, large by reality — is one of the more important findings in cognitive psychology. Most cognitive activities that feel demanding are demanding precisely because they push against the 3-to-5 ceiling.
Can working memory be trained?
This is the contested question and the basis of the entire "brain-training" industry.
The short, honest summary:
- Training on a specific working memory task reliably improves performance on that task. This is uncontroversial.
- Near transfer to closely related tasks is sometimes present, sometimes not. Depends on the specific training paradigm and the specific transfer task.
- Far transfer to general cognitive ability or real-world outcomes is not well supported by the evidence. The 2008 Jaeggi et al. paper claiming n-back training improves fluid intelligence has been mixed in replication; the 2016 Simons et al. Psychological Science in the Public Interest review concluded the far-transfer evidence was weak.
Senwitt is not built on broad-transfer claims about working memory training. See our does brain training work research page for the full position. What Senwitt does claim is the much narrower point that skills you actively use stay easier to use than skills you stop using — see skill atrophy for the long form.
Related Senwitt content
- The N-back test covers the working-memory task used most in fMRI and brain-training research.
- The digit span test covers the older, simpler working-memory measure used in clinical batteries.
- The Corsi block test is the spatial version.
- The memory skill page covers Senwitt's daily practice approach to active memory.
How the central-executive framing changed the field
Pre-1974, "short-term memory" was treated as a relatively passive storage buffer — a temporary holding tank for information on its way to long-term memory or out of the system. Baddeley and Hitch's working memory model reframed it as an active cognitive system with attentional control at its center.
The reframe was important because it explained patterns the old model couldn't. Why does mental arithmetic interfere with verbal rehearsal but not with spatial reasoning? Why do dual-task paradigms produce specific patterns of cost depending on which subsystems compete? Why does working memory capacity predict fluid reasoning better than short-term memory capacity? All these patterns sit naturally inside the Baddeley model and awkwardly outside it.
The Engle-Kane "executive attention" extension built on the same idea: working memory capacity is not really about storage at all; it's about attentional control applied to memory. The two views are now broadly compatible and dominate the contemporary working-memory literature.
A note on online working memory tests
Most "working memory test" sites online use either:
- A simplified N-back implementation, or
- A short backward-digit-span variant.
These are reasonable for getting a feel for working-memory load. They are not the research-grade complex-span tasks, and a single session is not a calibrated readout of your working memory capacity. Cognitive psychologists running validation studies use multi-session designs with multiple test variants because single-session estimates are noisy.
If you want to experience working-memory load — and the experience is genuinely informative, because the limits really are smaller than people expect — any of the online versions will get you there.
