Long-term memory (LTM) is the final, most durable stage of the multi-store memory model proposed by Atkinson and Shiffrin, holding information, skills, and experiences from minutes to a lifetime. It has no known limit on capacity or duration, and stores information mainly by meaning rather than sound.
Like a computer’s hard drive, LTM keeps information safely stored even when it isn’t currently on your desktop, or your short-term memory.
Key Takeaways
- Effectively Unlimited: Long-term memory has no known limit on capacity or duration. Most forgetting reflects a retrieval problem, not a lost memory trace.
- Availability vs Accessibility: A memory can be stored yet briefly hard to reach. The right cue often brings it back to mind.
- Two Main Types: Explicit memory (episodic and semantic) can be consciously recalled. Implicit memory, such as procedural skills, priming, and conditioning, shows itself through behaviour instead.
- Meaning-Based Encoding: Long-term memory mainly stores information by meaning rather than sound. Deep, meaningful processing builds stronger memories than rote repetition.
- Evidence From Amnesia: Patients such as HM could not form new declarative memories, yet they still learned new motor skills. This shows declarative and procedural memory rely on separate brain systems.
- Permastore: Bahrick’s classic study found people still recognised former classmates’ names and faces decades later, with some memories holding up after 48 years.
- Reconsolidation: Retrieving a memory can briefly destabilise it before it is “reconsolidated,” showing long-term memories are dynamic rather than fixed forever.
Types of Long-Term Memory
Long-term memory is not a single store and is divided into two types: explicit (knowing that) and implicit (knowing how).
One of the earliest and most influential distinctions of long-term memory was proposed by Tulving (1972), who separated episodic memory from semantic memory.
He added procedural memory as a third system only in 1985, arguing that episodic memory is embedded within semantic memory, which is itself embedded within procedural memory (Tulving, 1985).
Procedural Memory
Procedural memory is a part of the implicit long-term memory responsible for knowing how to do things, i.e., memory of motor skills.
It does not involve conscious (i.e., it’s unconscious-automatic) thought and is not declarative. For example, procedural memory would involve knowledge of how to ride a bicycle.
This works differently from remembering facts. Procedural memories are stored using a motor or action code rather than a verbal one. This is why skills are best learned by watching and practising, not by being told the steps.
Everyday examples include swimming, touch-typing, and tying shoelaces. A person can perform these fluently while finding it almost impossible to describe the exact sequence of movements in words. Procedural memories are also unusually durable, and they typically survive the amnesia that later devastates a person’s episodic and semantic memories.
Semantic Memory
Semantic memory is a part of the explicit long-term memory responsible for storing information about the world. This includes knowledge about the meaning of words, as well as general knowledge.
For example, London is the capital of England. It involves conscious thought and is declarative.
The knowledge that we hold in semantic memory focuses on “knowing that” something is the case (i.e. declarative). For example, we might have a semantic memory for knowing that Paris is the capital of France.
Episodic Memory
Episodic memory is a part of the explicit long-term memory responsible for storing information about events (i.e. episodes) that we have experienced in our lives.
It involves conscious thought and is declarative. An example would be a memory of our 1st day at school.
The knowledge that we hold in episodic memory focuses on “knowing that” something is the case (i.e. declarative). For example, we might have an episodic memory of knowing that we caught the bus to college today.
Priming and Conditioning
Priming is one form of implicit long-term memory. In priming, earlier exposure to a stimulus makes it easier to process a related stimulus later. There is no conscious memory of that earlier exposure (for example, seeing the word “doctor” speeds up recognising “nurse”).
The same applies to conditioning. Classical and operant conditioning create learned associations and emotional responses that shape behaviour automatically. A person can show a conditioned fear response with no conscious memory of the event that produced it.
Cohen and Squire (1980) drew a distinction between declarative knowledge and procedural knowledge.
Procedural and declarative knowledge map onto the implicit and explicit types described above. Procedural knowledge (“knowing how”, e.g. riding a bike or brushing your teeth) runs automatically with little conscious effort. Declarative knowledge (“knowing that”, e.g. that London is the capital of England) requires consciously bringing information to mind.
Evidence for this comes from amnesia research. Patients with amnesia typically have great difficulty retaining episodic and semantic information after the condition begins.
Their memory for events and knowledge acquired before the onset of the condition tends to remain intact. But they can’t store new episodic or semantic memories. In other words, it appears that their ability to retain declarative information is impaired.
However, their procedural memory appears to be largely unaffected. They can recall skills they have already learned (e.g. riding a bike) and acquire new skills (e.g. learning to drive).
A landmark example is patient HM (Henry Molaison), who lost the ability to form new episodic and semantic memories after surgery for epilepsy. He still learned new motor skills, such as mirror-drawing, despite having no conscious memory of ever practising them (Scoville & Milner, 1957).
Bahrick et al. (1975)
Aim: Bahrick et al. (1975) asked a simple question: can long-term memory last a lifetime? They studied what they called very long-term memory (VLTM).
Method: Nearly 400 participants aged 17–74 were tested on their memory for former high-school classmates. They completed a free recall test, a photo recognition test, and a name recognition test. The photo test used 50 pictures.
Results: Recognition held up remarkably well over time.
Within 15 years of leaving school, participants were about 90% accurate. Even after 48 years, they were still about 80% accurate for names and 70% for faces. Free recall was far weaker, falling from about 60% at 15 years to under 30% at 48 years.
Conclusion: Bahrick called this durable, plateau-like retention “permastore.” Some long-term memories, he concluded, can last essentially a lifetime. The gap between strong recognition and weak recall reflects an accessibility, not an availability, problem.
A major strength is the study’s high ecological validity: it tested memory for real, personally meaningful material rather than artificial word lists.
The design has real limits, though. Its cross-sectional approach could not fully separate the effect of time since learning from the effect of participant age. It also could not rule out rehearsal, such as later contact with old classmates, as a confound.
Critical Evaluation of Long-Term Memory Research
Long-term memory is a well-evidenced concept, but several of its central claims attract legitimate criticism. Four issues are worth noting:
- Untestable claims: “Unlimited” capacity and “permanent” duration cannot be proven directly, only inferred from the fact that no limit has ever been observed.
- A blurred episodic/semantic boundary: Semantic knowledge is often abstracted from episodic experience, so the two may not be fully separate systems.
- Systems versus process debate: Dissociations credited to separate memory systems could instead reflect different types of processing, not separate stores.
- Reconstruction, not reproduction: Episodic memories are rebuilt at recall, shaped by schemas and open to distortion, rather than played back intact.
Untestable Claims About Capacity and Duration
Neither claim can be tested directly. There is no way to prove a store has no upper bound, since that would mean confirming that no future item could ever fail to be stored.
“Permanent” duration faces the same problem. No experiment can run for an entire lifetime, so “permanent” can only be supported by an absence of counter-evidence. Even Bahrick’s finding that memories survive 48 years shows a long floor, not a proven ceiling.
The two can look identical.
Retrieval failure and genuine storage loss produce the exact same result from the outside. If someone cannot recall something, that alone cannot distinguish a decayed trace from one that is intact but temporarily inaccessible.
Every apparent failure can be reinterpreted as a retrieval problem. This makes the claim that nothing is ever truly erased close to unfalsifiable. Confident statements about what long-term memory categorically cannot do should therefore be treated with caution.
Is the Episodic/Semantic Split Genuinely Clean?
The dividing line is not always sharp in practice. Semantic knowledge is often thought to be gradually abstracted out of repeated episodic experience.
Many separate encounters with individual dogs, each an episodic memory, are eventually distilled into the general concept “dog.” The specific encounters themselves tend to be forgotten afterwards.
This process means episodic memory can feed into and become semantic memory over time, rather than the two remaining permanently separate stores.
Autobiographical memory makes the blur especially visible. Recalling your first day at school mixes a specific episodic scene with semantic facts about schools in general. It is not obvious which system is doing the remembering.
Cases like this can be argued either way. Some researchers therefore treat episodic and semantic memory as two points on a continuum within one declarative system, rather than as architecturally distinct systems.
Systems Versus Process Views
This article treats long-term memory as a set of distinct neural systems. Separate stores exist for episodic, semantic and procedural memory, each supported by characteristic dissociation evidence.
This systems view is associated most closely with theorists such as Tulving and Squire, and it remains the dominant framework in the field.
Process-based accounts offer a rival reading of the same evidence. Rather than crediting a dissociation to two separate stores, they argue it can reflect two different kinds of processing engaged by a task.
For example, conceptual processing, which draws on meaning, versus perceptual processing, which draws on surface features. A patient may lose one memory “type” but not another. On this view, that reflects a lost kind of processing, not an entire dedicated store.
The systems-versus-process disagreement remains unresolved for long-term memory, and it parallels a similar store-versus-process debate that runs through the short-term memory literature.
Episodic Memory Is Reconstructive, Not Reproductive
Two classic bodies of work make this point directly. Bartlett (1932), through his War of the Ghosts studies, showed that when people recall a story repeatedly over time, the account is not played back faithfully.
It is actively reconstructed, and unfamiliar or unexpected details tend to be revised toward what the rememberer already expects, guided by pre-existing schemas.
Loftus and Palmer (1974) showed the same reconstructive vulnerability experimentally. Simply changing the wording of a question asked after witnessing an event altered what people subsequently reported having seen.
Long-term memory should not be pictured as a video archive: what is retrieved is reconstructed at the moment of recall.
Recall is reconstructed, not replayed. This has consequences well beyond the laboratory. In eyewitness and forensic settings, misleading post-event information can distort what a witness later reports “remembering.”
This is why procedures such as the cognitive interview are designed to avoid introducing contaminating cues. Instead, they try to reinstate the original conditions under which the event was encoded.
Contemporary Research
Long-term memory is not a fixed archive. Modern neuroscience shows it can be updated, replayed, and even reshaped at the level of individual brain cells.
Memory Reconsolidation
Retrieving a stored memory can return it to a fragile, changeable state. The memory must then be “reconsolidated,” or restabilised, to survive.
Nader et al. (2000) found that blocking protein synthesis during this reconsolidation window weakens a reactivated fear memory in rats. This reframes long-term memory as dynamic rather than fixed, since a stored trace can be altered every time it is recalled.
This discovery has opened new treatment approaches for weakening maladaptive memories in conditions such as PTSD and addiction, by targeting the fragile window that follows retrieval.
Engram Cells
Liu et al. (2012) used optogenetics to reactivate the exact hippocampal neurons active during fear learning in mice. Stimulating these “engram cells,” the specific neurons thought to physically store a memory, triggered the fear memory directly, without the original cue.
This was the first direct evidence that a memory lives in an identifiable group of neurons. It is not distributed vaguely across the brain.
Related studies have gone further, implanting and reversing false memories in mice by manipulating the same engram cells. Together, this work gives a cellular-level account of both how memories are stored and how they can be distorted.
Sleep and Consolidation
During slow-wave sleep, the hippocampus “replays” newly learned information and gradually transfers it to the neocortex for lasting storage (Squire et al., 2015).
This gives a biological explanation for why sleep, rather than last-minute cramming, secures long-term learning, and it connects the classic consolidation debate to systems-level neuroscience.
Retrieval and Forgetting in Long-Term Memory
Long-term memory’s real limit is usually accessibility, not availability. How a memory is retrieved therefore matters as much as how it was stored, and it explains most of what looks like forgetting.
Encoding Specificity and Retrieval Cues
Tulving and Thomson’s (1973) encoding-specificity principle explains why. A retrieval cue works only when it overlaps with what was encoded at the time of learning. Memory is best when the conditions at retrieval reinstate those at encoding.
Thomson and Tulving (1970) demonstrated this directly. Recall was best when the cues at test matched those at study, and any change in cue reduced recall.
Even a shift from a weak cue at learning to a strong cue at test hurt recall. It had not been part of the original encoding.
This is also why recognition usually beats free recall. A recognition test supplies the studied item as a cue; free recall supplies none, exactly as Bahrick’s classmate study showed above.
The same principle explains context- and state-dependent memory. Retrieval improves when the external environment or internal mood at test matches that at encoding.
Interference, Decay and Retrieval Failure
Given long-term memory’s vast capacity, why do we forget? Most forgetting reflects retrieval failure, not the loss of a memory trace.
Interference is the best-supported cause of forgetting from long-term memory. Other memories simply compete with the target information. In proactive interference, older learning disrupts newer learning; in retroactive interference, newer learning disrupts older learning.
Decay is simply fading with time. It is well supported in short-term memory, but much harder to demonstrate in long-term memory. Apparent time-related loss in long-term memory is usually confounded with accumulating interference and changing retrieval cues, so its status here remains disputed.
The clearest evidence for retrieval failure is cue-dependent forgetting. A memory can feel completely lost, yet the right retrieval cue often brings it straight back.
This everyday tip-of-the-tongue experience is covered in full under theories of forgetting. Together, these mechanisms show why long-term memories so rarely vanish outright: what looks like forgetting is usually a retrieval problem waiting for the right cue.
References
Bahrick, H. P., Bahrick, P. O., & Wittinger, R. P. (1975). Fifty years of memory for names and faces: A cross-sectional approach. Journal of Experimental Psychology: General, 104(1), 54–75.
Cohen, N. J., & Squire, L. R. (1980). Preserved learning and retention of pattern-analyzing skill in amnesia: Dissociation of knowing how and knowing that. Science, 210(4466), 207–210. https://doi.org/10.1126/science.7414331
Liu, X., Ramirez, S., Pang, P. T., Puryear, C. B., Govindarajan, A., Deisseroth, K., & Tonegawa, S. (2012). Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature, 484(7394), 381–385.
Nader, K., Schafe, G. E., & LeDoux, J. E. (2000). Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval. Nature, 406(6797), 722–726.
Scoville, W. B., & Milner, B. (1957). Loss of recent memory after bilateral hippocampal lesions. Journal of Neurology, Neurosurgery, and Psychiatry, 20(1), 11–21.
Squire, L. R., Genzel, L., Wixted, J. T., & Morris, R. G. (2015). Memory consolidation. Cold Spring Harbor Perspectives in Biology, 7(8), a021766.
Thomson, D. M., & Tulving, E. (1970). Associative encoding and retrieval: Weak and strong cues. Journal of Experimental Psychology, 86(2), 255–262.
Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381–403). Academic Press.
Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26(1), 1–12.
Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80(5), 352–373.
