Open any introductory psychology textbook and you will find the same diagram: two boxes, one labeled “episodic memory” for personal life events, the other labeled “semantic memory” for general facts and knowledge. The split has been taught as settled science since Endel Tulving first proposed it in 1972. Now a brain-imaging study led by Robert Sheridan at the University of Nottingham, published in May 2026 as a Registered Report in Nature Human Behaviour, reports that when task demands are carefully equalized, retrieving a fact and retrieving a life event light up nearly identical brain networks. Bayesian statistics did not just fail to find a difference; they actively favored the conclusion that no real neural difference exists.
If the finding survives replication, it challenges one of the most durable ideas in cognitive science and could reshape how clinicians evaluate memory disorders from Alzheimer’s disease to post-traumatic amnesia.
What the Nottingham team actually did
Sheridan and colleagues used a multi-echo fMRI protocol to scan 40 volunteers while they performed tightly controlled retrieval tasks built around logo-and-brand pairings. In the semantic condition, participants drew on real-world knowledge they already had (matching a logo to the brand it belongs to). In the episodic condition, they retrieved brand-logo associations that had been newly taught to them in the lab, simulating the encoding of a personal experience. Stimuli, response demands, and difficulty levels were matched across conditions so that any difference in brain activation could be pinned to the type of memory, not to unrelated task features.
The central result: no reliable difference in neural activity between successful semantic and episodic retrieval. Bayes factors in the range of 3 to 10 supported the null hypothesis across key brain regions, meaning the data provided positive evidence that the two memory types share the same retrieval machinery. That is a stronger statistical claim than a conventional non-significant result, which would simply leave the question open. “We expected to find at least some regions that clearly preferred one memory type over the other,” Sheridan said in the university’s press release. “Instead, the Bayesian evidence consistently pointed toward equivalence.”
Crucially, the study was conducted as a Registered Report. That means the hypotheses, sample size, and full analysis plan were peer-reviewed and locked in before a single scan was run. This format exists specifically to prevent the kind of post-hoc analysis shopping that has inflated false-positive rates across neuroimaging research. It gives the null result considerably more weight than a standard exploratory fMRI paper would carry.
Why the episodic-semantic divide became orthodoxy
The distinction the Nottingham study targets traces back to a 1972 book chapter in which Tulving formally argued that episodic and semantic memory are separate cognitive systems. That theoretical split gained clinical weight from patient H.M. (later identified as Henry Molaison), whose bilateral medial temporal lobe resections left him unable to form new episodic memories while preserving much of his general knowledge. For decades, the H.M. case and similar lesion studies were read as proof that distinct brain circuits handle facts and personal events.
That reading shaped everything from undergraduate curricula to clinical neuropsychological test batteries, many of which still categorize certain questions as “purely semantic” or “purely episodic” when scoring patients.
Converging evidence from a different technology
A separate line of evidence points in the same direction. A 2026 EEG preprint from an independent research group identified what its authors describe as a shared retrieval state engaged during both episodic and semantic recall. Because EEG measures electrical activity on a millisecond timescale while fMRI tracks blood-flow changes over seconds, convergence across the two methods strengthens the case that the overlap is not simply an artifact of one imaging technique’s limited resolution.
That said, the two studies used different participants, different stimuli, and different analytic frameworks. Their convergence is suggestive, not definitive. Formal cross-validation linking the fMRI activation maps to the EEG retrieval-state signatures has not been published, and preprints have not completed full peer review. Treating the two as independent, converging signals is reasonable; treating them as a single unified proof would overstate the evidence.
What the study cannot tell us yet
Forty participants is a respectable but moderate sample for whole-brain fMRI. The specific trial-by-trial behavioral accuracy rates are described at a high level in the university’s press materials; independent spot checks by other researchers will depend on full examination of the raw data hosted on the Open Science Framework.
The bigger question is generalizability. Logo-brand pairings are a narrow slice of what people ordinarily mean by “facts” or “life events.” Recalling where you stood during your wedding, for instance, involves rich sensory detail, emotional charge, and a sense of personal time that a brand-logo task does not capture. Whether the neural overlap holds for more vivid, autobiographical memories is something the current data simply cannot answer.
Lesion and patient data also complicate the picture. The classic observation that hippocampal damage can devastate episodic memory while leaving semantic knowledge partly intact has been replicated many times. If both memory types truly rely on the same networks, the selective vulnerability of episodic memory in amnesia patients demands a new explanation. One possibility raised by researchers in the field is that the difference lies not in which brain regions are recruited but in how those regions coordinate timing or information flow, a distinction that fMRI’s relatively coarse temporal resolution may not capture.
What this means for textbooks and clinics
For students, educators, and clinicians who have built their understanding of memory around the episodic-semantic divide, the emerging picture calls for an update, not a demolition. Textbooks may need to shift from portraying two cleanly separated systems to describing partially overlapping processes that can be teased apart under some conditions but not others. Clinicians who rely on standard assessments premised on a sharp distinction, such as tests that label certain questions as “purely semantic,” may need to interpret results with more caution when drawing inferences about which brain structures are impaired.
The Nottingham Registered Report does not prove that episodic and semantic memory are identical in all circumstances. What it does is demonstrate, with a level of methodological rigor uncommon in neuroimaging, that the expected neural dissociation vanishes when retrieval tasks are properly matched. That is a meaningful crack in a 54-year-old wall.
Where replication and richer paradigms will be decisive
Studies that use richer autobiographical events, vary emotional intensity, or track memory over longer delays could reveal conditions under which episodic and semantic retrieval diverge more clearly in the brain. Combining fMRI with methods that capture neural timing, such as EEG or magnetoencephalography, may help determine whether the apparent overlap in activated regions masks subtler differences in information flow.
For now, one of cognitive neuroscience’s foundational assumptions has met a serious, well-designed challenge. The traditional story of separate systems for facts and events is no longer the only viable framework. Instead, the evidence is tilting toward a more integrated view of memory, one in which what we know and what we remember draw on a common neural infrastructure that can be configured differently depending on context. Whether that integrated view becomes the new textbook standard or settles in as one piece of a more complex theory will depend on the datasets that follow.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
