Long-term memory is broadly divided into declarative (explicit) and non-declarative (implicit) memory. Declarative memory encompasses episodic memory (personally experienced events with temporal-spatial context) and semantic memory (general world knowledge, concepts, and facts). Non-declarative memory includes procedural skills, priming, and classical conditioning. These systems are dissociable: amnesic patients can show intact procedural learning despite profound declarative memory loss.
Study the famous case of H.M. (Henry Molaison) — he could learn new motor skills (procedural) but could not form new episodic memories. Contrasting amnesic versus intact populations reveals the architecture of long-term memory systems.
From your study of the working memory model, you have a picture of how information is held temporarily in the phonological loop, visuospatial sketchpad, and episodic buffer, with the central executive directing attention. Working memory is a workspace — powerful but severely limited in capacity and duration. Long-term memory is its complement: a vast, persistent store whose capacity appears effectively unlimited. The central question in this topic is not "how much does long-term memory hold?" but "what kinds of things does it hold, and how does it hold them?" The answer turns out to be: multiple distinct systems with different contents, different neural substrates, and different access routes.
The highest-level division is between declarative (explicit) memory and non-declarative (implicit) memory. Declarative memory contains knowledge that can be consciously accessed and stated — declared in words or images. It subdivides into episodic memory (memories with autobiographical context: *when* and *where* something happened to *you*, experienced as mental time travel) and semantic memory (context-free factual knowledge: that Paris is the capital of France, that dogs are mammals). The episodic/semantic distinction matters because these systems can be selectively damaged: semantic dementia can strip away world knowledge while leaving autobiographical memory relatively intact, and certain amnesic patients show the reverse.
Non-declarative memory is not a unified system but a family of systems that share the property of being expressed through *performance* rather than deliberate recollection. Procedural memory underlies skills — riding a bike, typing, playing piano. You perform them without recalling how you learned them; the knowledge is in the doing, not the telling. Priming is the facilitation of processing by prior exposure to a stimulus, even without conscious recognition — seeing "nurse" processes faster after "doctor" because the prior encounter left a trace. Classical conditioning is also implicit: the conditioned fear response to a tone paired with shock is a learned association expressed automatically, not through deliberate recall.
The most compelling evidence for these distinctions comes from neuropsychological dissociation, particularly the patient H.M. (Henry Molaison). Following bilateral hippocampal removal for epilepsy, H.M. was profoundly amnesic for new declarative memories — he could not remember conversations from minutes earlier, could not learn new facts, and lived in a perpetual present. Yet he could learn new motor skills: across sessions of mirror-drawing practice, his performance improved normally, even though each day he had no recollection of having practiced before. His procedural system was intact; his declarative system was destroyed. This dissociation confirmed that long-term memory is not a single faculty but an architecture of distinct systems with different neural homes — the hippocampus is critical for declarative memory, while the basal ganglia, cerebellum, and amygdala support different forms of non-declarative memory independently.