A new fMRI study found that the brain activity produced when a person recalls a memorized fact and when that same person relives a personal experience is nearly indistinguishable at the network level. The research, which directly compared episodic and semantic retrieval within the same subjects during a single scanning session, found no reliable differences in either predefined brain networks or task-driven activation clusters. The result challenges a decades-old assumption in cognitive neuroscience that these two forms of memory depend on separable neural machinery.
Why the overlap between fact recall and event memory matters right now
For more than 50 years, textbooks have drawn a bright line between episodic memory, the system that lets a person mentally travel back to a birthday party or a first day at work, and semantic memory, the system that stores general knowledge such as the capital of France or the color of a fire truck. That distinction, first proposed by Endel Tulving in 1972, shaped how clinicians diagnose memory disorders, how educators design curricula, and how drug trials measure cognitive decline. If the two systems share far more neural real estate than previously believed, each of those applications needs re-examination.
The new study used a clever within-subject design. Participants first learned novel logo–brand pairings and then entered the scanner, where they retrieved both those freshly encoded associations (episodic retrieval) and their pre-existing real-world knowledge of the same brands (semantic retrieval). According to work reported in Nature Human Behaviour, neither a priori networks nor task-activated clusters showed evidence for differences between the two retrieval types when both were successful. That null result held across multiple analytical approaches, making it difficult to dismiss as a statistical fluke.
One plausible explanation centers on retrieval fluency. When a person pulls up a fact quickly and confidently, the cognitive demands may mirror those of smoothly reliving a vivid event. High-fluency semantic trials could show even greater spatial correlation with episodic activation patterns than low-fluency ones, suggesting that what drives neural overlap is not memory type but the ease and depth of the retrieval process itself. Testing that hypothesis in future work would require trial-by-trial reaction-time data paired with voxel-level activation maps, a design the current study’s framework could support.
Converging fMRI evidence for shared retrieval networks
The Nature Human Behaviour finding did not emerge in isolation. A separate fMRI project in Nature Communications showed that both semantic and episodic retrieval recruit overlapping brain systems and that shared processes, including cognitive control demands, can drive that overlap. By varying how much executive control each trial required, the researchers demonstrated that the degree of network convergence tracked with control load rather than with memory category, implying that the brain may organize retrieval more around process than content labels.
A scholarly review aggregating data across multiple experiments identified extensive shared regions spanning midline frontal cortex, middle temporal gyrus, parahippocampal cortex, ventral parietal cortex, and midline posterior regions for both personal semantic and episodic recall, according to a full-text synthesis in PubMed Central. Those regions overlap heavily with the brain’s default network, a set of areas active during internally directed thought, daydreaming, and remembering the past.
An earlier meta-analysis published in Neuropsychologia also reported extensive, largely overlapping default network regions for episodic and semantic retrieval. That analysis did, however, identify one area of divergence: hippocampal and parahippocampal regions contributed more to episodic retrieval than to semantic retrieval. This creates a direct tension with the newer within-subject data, which found no such distinction at the network level.
Hippocampal divergence and the limits of current data
The conflict between these findings deserves close attention. The Neuropsychologia meta-analysis, which pooled results across many separate experiments with different designs and participant pools, flagged hippocampal and parahippocampal involvement as the clearest separator of episodic from semantic memory. Yet the Nature Human Behaviour study, using a tightly controlled within-subject comparison, found that even those regions did not reliably distinguish the two retrieval types.
Several factors could explain the discrepancy. Meta-analyses combine studies that use different baseline tasks, stimulus types, and scanning parameters, which can amplify small regional differences into apparent categorical distinctions. A within-subject design eliminates between-subject variability and controls for task context more precisely, potentially washing out differences that are artifacts of experimental heterogeneity rather than genuine signatures of separate memory systems.
Task difficulty may also matter. If episodic trials in earlier work were systematically harder, slower, or less accurate than semantic trials, hippocampal engagement could partly reflect effort, reconstruction, or error monitoring rather than a unique episodic code. In contrast, when both forms of retrieval are equated for success and fluency, as in the logo–brand paradigm, hippocampal responses may converge.
Another possibility is that the hippocampus differentiates memories at a finer grain than the broad episodic–semantic labels allow. The same hippocampal circuitry might support detailed relational processing-binding together people, places, and objects-whenever that is required, regardless of whether the memory is classified as a personal episode or a well-learned fact. On this view, the observed overlap is not a failure to find the “right” contrast, but a clue that the traditional taxonomy is misaligned with neural function.
Rethinking memory categories and clinical practice
A separate Nature Communications report on distinct cortical systems for content and context in episodic memory suggests that the critical division may lie within episodic memory itself, between what happened and where or when it happened, rather than between episodic and semantic memory as wholes. If networks track content versus context, or control demand versus automatic retrieval, then decades of interpreting activation maps through the episodic–semantic lens may need revisiting.
For clinicians, the implications are significant. Diagnostic tools often assume that episodic memory loss, as in early Alzheimer’s disease, can be cleanly separated from preserved semantic knowledge. If the same large-scale networks support both, early degeneration might degrade factual knowledge and personal experiences in parallel, with apparent selectivity emerging only because certain tasks are more demanding. Neuropsychological batteries may need to emphasize process characteristics-such as reliance on controlled search or relational binding-rather than nominal memory type.
In education, the findings challenge the idea that teaching strategies can target “episodic” versus “semantic” systems independently. Techniques that promote richer encoding, multiple retrieval attempts, and varied contexts may strengthen a common retrieval infrastructure that serves both storytelling and fact recall. The boundary between learning a concept and remembering a specific example of it could be more fluid than curriculum designers typically assume.
For basic science, the converging evidence argues for models that treat memory as an interplay of overlapping networks tuned by task goals, control demands, and representational detail. Instead of asking where episodic memory “lives,” researchers may gain more traction by mapping how control regions, default-network hubs, and medial temporal structures cooperate across different kinds of remembering.
The new within-subject data do not erase all distinctions between remembering events and recalling facts, but they do undermine the notion that the brain maintains two neatly separable systems. As more studies adopt designs that equate difficulty, success, and stimulus content across conditions, the field may move toward a more process-based taxonomy of memory-one in which the same networks flexibly support both a childhood birthday and the capital of France, depending on what the moment demands.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
