Associative memory
Knowledge stored as a web of linked ideas, where activating one pulls up its neighbors.
What it means
Associative memory is the storage and retrieval of information as a network of connected elements, such that encountering or thinking of one item automatically makes related items more accessible. On this view memory is not a set of isolated files but a web of nodes—concepts, words, images, feelings—joined by links whose strength reflects how often and how meaningfully they have co-occurred. Retrieval works by spreading activation: energizing one node sends activation outward along its links, priming associated nodes and making them faster to recognize or recall. This architecture explains priming, the way a cue evokes a cascade of related thoughts, and the constructive, sometimes error-prone nature of recall, since strongly associated but never-presented items can feel like genuine memories. It is the substrate Kahneman invokes for fast, intuitive System 1 thinking, where a coherent story assembles itself from whatever associations a situation triggers. It matters because so much of judgment runs on which associations a context happens to activate.
How activation spreads
Activation is graded and it fades. A cue energizes its node, and the charge travels outward along links, weakening at every step and decaying rapidly unless something refreshes it. Link strength decides how far it reaches: strong neighbours light up fully, remote ones barely register. This is why priming is measured in tens of milliseconds rather than in whole thoughts. It shifts what is accessible, not what you believe. The same geometry predicts mediated priming: in naming tasks 'lion' speeds 'stripes' via the unmentioned 'tiger' — though the effect vanishes in lexical decision, where a post-retrieval relatedness check appears to swamp it. Timing matters too: at short intervals the spread is automatic, but given more time, deliberate strategies layer on top and can mask it.
What the evidence shows
The core demonstration is the lexical decision task: people confirm that 'nurse' is a word faster when 'doctor' precedes it than when an unrelated word does. Meyer and Schvaneveldt reported this in 1971 and it has held up. The largest test to date measured priming across 19 languages with more than 25,000 participants and found the effect present throughout, though variable enough that models fit better once each language was allowed its own intercept. Lucas's meta-analysis showed priming survives without any associative link, since shared features alone produce it, while adding an association gives a further boost. Roediger and McDermott's list paradigm shows the same machinery generating errors: people falsely recalled the unpresented associate 40 to 55 percent of the time, with confidence.
A competing explanation
The network picture is the standard one and it earns its keep, but spreading activation is a metaphor, and not the only one that fits the data. Ratcliff and McKoon proposed instead that prime and target combine into a single compound cue matched against memory as a whole; related pairs return a stronger familiarity signal, and nothing needs to travel anywhere. Reaction times rarely separate the two accounts, since both predict that related pairs are handled faster. The distinction matters more than it looks: if you picture literal energy running down wires, you will over-predict long chains and distant effects. The safer claim is the one both models share, that recent context changes what memory hands back, quickly and without asking.
Limits and caveats
Priming is robust in aggregate and unreliable per person. Stolz, Besner and Carr found that an individual's priming score barely agrees with itself across sessions: the group effect is real, while any one person's number is mostly noise. It cannot serve as a trait measure. Keep semantic priming apart, too, from the behavioural priming claims that collapsed under replication in the 2010s. That a word speeds recognition of a related word is well established; that reading words about old age slows your walking is not. Associative accessibility is a small, fast, local effect, and most of the trouble comes from treating it as a lever on behaviour.
Using it in practice
Treat surrounding context as part of any stimulus you present, because it decides which associations arrive first. In measurement this means order effects are not a nuisance to wave away: an earlier item leaves residue on a later one, so randomize order or hold it fixed and report which you did. In design it means the words and images sitting near a decision quietly stock the shelf people reach from. The honest use is defensive rather than manipulative. Before an important judgment, ask what was activated beforehand and whether it has any business being relevant. That question catches more errors than any attempt to prime someone on purpose.
Examples
Hearing the word 'doctor' speeds recognition of 'nurse,' and studying a list of sleep-related words can leave people falsely certain they saw the word 'sleep.'
A brand plays the same jingle over beach scenes for years; on a hot day the tune surfaces unbidden and the drink somehow feels like the obvious thing to want.
Walk into your childhood kitchen and the smell pulls up the dog, the wallpaper, an argument you had forgotten — one cue lights a whole neighbourhood of memory at once.
A clinician who has just worked through three flu cases meets a patient with fever and aches. The flu network is already active, and the one atypical detail pointing elsewhere never becomes salient.
A survey asks about neighbourhood crime, then about life satisfaction. The crime item leaves threat-related associations active, and life-satisfaction ratings shift — which direction depends on whether respondents fold the crime item into the global judgment or set it aside as already answered.
Key references
- Buchanan, E. M., Cuccolo, K., Heyman, T., et al. (2026). Measuring the semantic priming effect across many languages. Nature Human Behaviour, 10(1), 182-201. doi.org/10.1038/s41562-025-02254-x
- Stolz, J. A., Besner, D., & Carr, T. H. (2005). Implications of measures of reliability for theories of priming: Activity in semantic memory is inherently noisy and uncoordinated. Visual Cognition, 12(2), 284-336. doi.org/10.1080/13506280444000030A
- Lucas, M. (2000). Semantic priming without association: A meta-analytic review. Psychonomic Bulletin & Review, 7(4), 618-630. doi.org/10.3758/BF03212999
- Roediger, H. L., & McDermott, K. B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21(4), 803-814. doi.org/10.1037/0278-7393.21.4.803
- Ratcliff, R., & McKoon, G. (1988). A retrieval theory of priming in memory. Psychological Review, 95(3), 385-408. doi.org/10.1037/0033-295X.95.3.385
- Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82(6), 407-428. doi.org/10.1037/0033-295X.82.6.407