In 1972, the Estonian-Canadian psychologist Endel Tulving at the University of Toronto proposed that long-term memory was not one thing but two. Semantic memory held general knowledge — Paris is the capital of France; water boils at 100 °C — facts unattached to any particular moment of learning them. Episodic memory held specific personal events — what you ate for breakfast yesterday; where you were on September 11 — temporally and spatially bound, accompanied by a sense of re-experiencing. The distinction was philosophical in 1972 and empirical by the 1990s, when amnesic patients and neuroimaging both showed the two memory types could dissociate cleanly. The hippocampus turned out to be essential for forming new episodic memories; semantic memory was widely distributed in the neocortex. Tulving had identified one of the brain's most fundamental architectural divisions.
Declarative memory — the kind you can describe verbally — is distinguished from non-declarative memory (procedural skills, classical conditioning, priming) by both neural substrate and phenomenology. Episodic memory carries autonoetic consciousness — Tulving's term for the sense of subjectively re-living an event — spatial-temporal context, first-person perspective, and high but fading vividness; it is reconstructive and so vulnerable to distortion. Semantic memory by contrast is noetic — knowing without re-experiencing — has no temporal anchor, is fact-like in format, and is relatively stable. The most studied case is patient H.M. — Henry Molaison — who had bilateral medial temporal resection in 1953 to control intractable epilepsy and as a result could form no new episodic memories for the rest of his life (until his death in 2008), yet retained normal semantic knowledge acquired before the surgery, normal short-term memory, normal procedural learning, and normal intelligence. Subsequent patients refined the picture: episodic memory is hippocampus-dependent always, while adult semantic knowledge is stored primarily in neocortical networks. The standard systems-consolidation model (Squire) holds that new memories are initially hippocampus-dependent and gradually become hippocampus-independent over weeks to years, though retrograde amnesia gradients are flatter than the model predicts — multiple-trace theory (Nadel and Moscovitch) holds that episodic detail remains hippocampus-dependent permanently. Imaging confirms the dissociation: episodic encoding and retrieval activate the hippocampus and medial temporal lobe, while semantic retrieval activates a distributed left-hemispheric network. Semantic dementia — focal neurodegeneration of the anterior temporal lobes — produces progressive loss of semantic knowledge while episodic memory is initially preserved, the inverse of H.M.'s dissociation. Episodic future thinking uses substantially the same neural circuitry as episodic memory; amnesics are also impaired at imagining the future.
Episodic memory is the memory most affected by normal aging and earliest in Alzheimer's disease — the entorhinal cortex is among the first regions to show pathology — while semantic memory is relatively preserved until late stages. Trouble remembering recent events while still accessing general knowledge is a diagnostic hallmark of early Alzheimer's. Eyewitness testimony is episodic memory under high stakes, and its reconstructive nature makes it systematically unreliable. The Innocence Project has documented hundreds of wrongful convictions overturned by DNA evidence, with mistaken eyewitness identification the leading cause. Trauma and PTSD involve episodic memories that fail to consolidate normally; prolonged exposure therapy and EMDR exploit reconsolidation to dampen them. Spaced-repetition systems (Anki, Duolingo) target semantic memory, exploiting the spacing effect first measured by Ebbinghaus in 1885. AI memory architectures in large language models are essentially semantic — generalizations stored without episodic context — and retrieval-augmented generation attempts to graft an episodic layer onto the semantic substrate.