Declarative knowledge is knowing that something is the case, while procedural knowledge is knowing how to do something. You know that Paris is the capital of France (declarative), and you know how to ride a bicycle (procedural). This distinction, rooted in the philosophical work of Gilbert Ryle and later formalized in cognitive psychology, turns out to be more than a neat conceptual split. The two types of knowledge depend on different brain circuits, develop on different timelines, respond differently to disease, and break down in surprisingly independent ways.
Two Memory Systems, Two Brain Networks
The reason declarative and procedural knowledge feel so different is that they literally live in different parts of your brain. Declarative memory, the system responsible for storing facts and personal experiences, depends heavily on structures in the medial temporal lobe, including the hippocampus and surrounding cortex. Damage to both sides of this region leads to problems forming and retrieving memories for facts and events.1Neurobiology of Learning and Memory. The medial temporal lobe and memory When you recall a friend’s birthday or remember what you had for lunch yesterday, these temporal-lobe structures are doing the heavy lifting. The hippocampal region, for instance, shows greater activity when you successfully encode associative information, like linking a new face to a name, compared to when you just process isolated details.2PubMed Central. Medial temporal lobe activation during encoding and retrieval of novel face-name pairs
Procedural memory, by contrast, relies on a different network. Motor skills and learned sequences depend on circuits connecting the cortex to the basal ganglia and the cerebellum. Brain imaging studies of people learning motor skills show that these two sub-networks contribute in distinct ways: the cortico-striatal system (linking the cortex to the basal ganglia) is more involved in learning sequences of movements, while the cortico-cerebellar system plays a larger role in adapting movements to new conditions.3Neuropsychologia. Distinct contribution of the cortico-striatal and cortico-cerebellar systems to motor skill learning This separation is not absolute — the two networks cooperate, and some tasks draw on both. But the core architecture means that losing one system does not necessarily cripple the other, a fact that becomes vivid when we look at neurological disease.
How Knowing That Becomes Knowing How
Think about learning to drive a car. Early on, you are acutely aware of every step: check the mirrors, ease off the clutch while pressing the accelerator, judge the gap before merging. You could describe each action in words. This is your declarative system working overtime, supplying rules and verbal instructions. Months later, you do all of it without thinking. That shift from effortful, conscious control to smooth, automatic execution is the hallmark of a skill transitioning from declarative to procedural knowledge.
The most influential model of this process, proposed by Paul Fitts and Michael Posner in the 1960s, breaks skill learning into three stages. In the cognitive stage, you think about what to do and make many errors. In the associative stage, you start linking the right actions together and errors drop. In the autonomous stage, the skill runs on autopilot and you can hold a conversation or think about something else entirely. This framework has held up across domains. In surgical training, for example, experienced surgeons display significantly more automaticity when performing simulated brain-tumor resections than novices do, consistent with the Fitts and Posner model.4PubMed. Automaticity of Force Application During Simulated Brain Tumor Resection: Testing the Fitts and Posner Model Recent neuroimaging work has even used directed functional connectivity between brain regions as a biomarker to track people’s progression through these stages, confirming that the neural signature genuinely changes as a learner moves from effortful to automatic performance.5arXiv. Beyond Performance Scores: Directed Functional Connectivity as a Brain-Based Biomarker for Motor Skill Learning and Retention
One important update to this classic picture: the transition is not as clean as Fitts and Posner originally implied. Rather than an orderly handoff from strategy-based (declarative) control to automatized (procedural) control, research now suggests the two processes operate with considerable independence throughout learning. The balance shifts — you rely less on conscious strategy and more on automatic execution over time — but strategic and adaptive processes do not simply switch off once automaticity kicks in.6PubMed Central. The role of strategies in motor learning That coexistence matters, as we will see when we look at what happens under pressure.
When Automaticity Backfires
If procedural knowledge runs on autopilot, you might expect that experts would be immune to pressure. They have done the thing ten thousand times; what is there to choke on? The answer, according to research on skilled performance under stress, is that pressure can cause you to switch off your procedural system and turn your declarative system back on. This is sometimes called the explicit monitoring hypothesis: when the stakes feel high, you start paying conscious attention to movements that normally run without your supervision, and that extra attention disrupts the smooth execution your procedural system had already mastered.7PubMed Central. Choking and Excelling Under Pressure in Experienced Classifiers
This is the paradox of choking: the very knowledge that helped you learn the skill in the first place (explicit rules, conscious focus) becomes a liability once the skill is proceduralized. A basketball player who starts thinking about the mechanics of a free throw mid-game, or a musician who suddenly becomes conscious of finger placement, can fall apart precisely because they are trying to use the wrong type of knowledge for the task. The practical takeaway for performers is counterintuitive — under pressure, you often want to distract your declarative system rather than engage it. Counting backward, humming, or focusing on an external target can keep you from accidentally micromanaging a skill that runs better on its own.
Language and the Declarative-Procedural Split
The distinction between these two knowledge types extends well beyond motor skills. One of the more interesting applications is in language. The declarative/procedural model of language, developed by neuroscientist Michael Ullman, proposes that your mental dictionary of memorized words depends on the same temporal-lobe declarative memory system that stores facts and events, while your mental grammar — the rules for combining words into phrases and sentences — depends on the frontal-basal-ganglia-cerebellar network that underlies procedural memory.8PubMed. The declarative/procedural model of lexicon and grammar
This is not just a theoretical nicety. The model makes testable predictions. If vocabulary relies on declarative memory and grammar relies on procedural memory, then brain damage that impairs one system should affect the corresponding language ability while leaving the other relatively intact. And that is broadly what the evidence shows. Patients with temporal-lobe damage often struggle to recall words but can produce grammatically correct sentences. Patients with basal-ganglia damage, as in certain forms of Parkinson’s disease, show the opposite pattern. The model also extends to second-language learning: adults learning a new language tend to rely more heavily on declarative memory at first (memorizing vocabulary and grammar rules as explicit facts), while the grammar of their native language runs more procedurally.9PubMed. Contributions of memory circuits to language: the declarative/procedural model
What Alzheimer’s and Parkinson’s Reveal
Neurological diseases provide some of the clearest evidence that declarative and procedural knowledge are dissociable. Alzheimer’s disease attacks the medial temporal lobe early, devastating declarative memory: people lose the ability to form new memories of events and facts, forget names and faces, and become disoriented in time. But their procedural memory can remain surprisingly intact. A systematic review and meta-analysis of studies on procedural learning in people with amnestic mild cognitive impairment or Alzheimer’s dementia found that the difference in procedural learning between these groups and healthy older adults was not statistically significant and was smaller than the threshold for even a trivial effect.10PubMed Central. Procedural Learning in Individuals with Amnestic Mild Cognitive Impairment and Alzheimer’s Dementia: a Systematic Review and Meta-analysis In practical terms, someone with early-to-moderate Alzheimer’s may not remember learning a skill but can still improve at it through practice. This has real implications for rehabilitation and daily living: procedural routines can sometimes compensate for declarative losses.
Parkinson’s disease tells the story from the other side. Because Parkinson’s primarily involves the degeneration of dopamine-producing neurons in the basal ganglia, you would expect procedural learning to be the harder hit, and it is. Research comparing Parkinson’s subtypes found that patients whose dominant symptom was slowness of movement (bradykinesia) were significantly impaired on procedural learning tasks compared to controls, while patients whose dominant symptom was tremor were not.11PubMed. Declarative and procedural learning in Parkinson’s disease patients having tremor or bradykinesia as the predominant symptom Both subgroups performed similarly on declarative tasks. Other research shows that Parkinson’s can impair both memory types, but the procedural deficit is specifically tied to the disease pathology rather than simply to aging.12PubMed. Procedural memory and Parkinson’s disease The pattern across these diseases reinforces the two-system architecture: damage the temporal lobe and declarative memory suffers; damage the basal ganglia and procedural learning suffers.
How Aging Affects the Two Systems Differently
Even without disease, normal aging does not treat declarative and procedural knowledge equally. A study comparing younger and older adults on both types of tasks found a clear dissociation. On declarative tasks like word lists and visual pair associations, younger adults had a steeper learning curve, picking up the material faster. But on procedural tasks like puzzle-solving and maze-navigation, the learning rate of both groups was similar, and in one case the older group actually improved faster than the younger group.13The Journals of Gerontology: Series B. Baseline Performance and Learning Rate of Procedural and Declarative Memory Tasks: Younger Versus Older Adults
This is genuinely good news for older adults and for anyone designing programs to help them stay independent. If your procedural learning apparatus holds up better than your declarative one, then building skills through repeated practice, rather than relying on memorized instructions, may be a more effective strategy. It also helps explain why older adults can remain highly competent at complex physical tasks (driving, playing an instrument, cooking elaborate recipes) even as they notice their memory for names and recent events declining.
Teaching to Both Systems
The declarative-procedural distinction has practical consequences for how people should be taught. In mathematics education, researchers distinguish between declarative knowledge (knowing that 7 × 8 = 56), procedural knowledge (knowing how to carry out long division), and conceptual knowledge (understanding why a procedure works). Evidence-based practice emphasizes the importance of balancing all three knowledge types across the math curriculum.14Learning Disabilities Research & Practice. Using Evidence–Based Practices to Build Mathematics Competence Related to Conceptual, Procedural, and Declarative Knowledge A student who has memorized multiplication facts (declarative) but cannot apply a multi-step problem-solving procedure (procedural) is only partially equipped. Conversely, a student who can follow a procedure by rote but does not understand the underlying concepts will struggle when the problem changes shape.
Medical education has explored similar ground. A study on case-based blended learning for medical students found that combining e-learning case presentations with practice improved performance outcomes, effectively helping students bridge the gap between knowing facts about a disease and knowing how to manage patients with that disease.15PubMed Central. Does case-based blended-learning expedite the transfer of declarative knowledge to procedural knowledge in practice? In other words, the most effective educational approaches deliberately move learners from declarative knowledge (reading about symptoms, memorizing drug doses) to procedural knowledge (diagnosing patients, managing emergencies) rather than hoping the transition happens on its own.
Research on students with ADHD adds another dimension. When ADHD students were taught using hypermedia instruction instead of traditional lectures, the benefits were strongest for procedural knowledge and showed up mainly in retention, meaning the students remembered how to do things better over time.16PubMed. Effects of hypermedia instruction on declarative, conditional and procedural knowledge in ADHD students The implication is that the medium of instruction matters, and it may matter differently depending on which type of knowledge you are trying to build.
Drugs and the Two Systems
Pharmacology provides another window into the independence of declarative and procedural knowledge. In a study of healthy volunteers, the drug D-cycloserine (which enhances a type of glutamate receptor activity) facilitated procedural but not declarative learning. Interestingly, when valproic acid was given alongside it, the procedural boost disappeared, suggesting that the two drugs had opposing effects on the neural circuits underlying skill learning.17PubMed. D-cycloserine facilitates procedural learning but not declarative learning in healthy humans
A broader pharmacological study tested the effects of three different classes of drugs on both memory systems: haloperidol (which blocks dopamine), midazolam (which boosts GABA activity, the brain’s main inhibitory signal), and scopolamine (which blocks acetylcholine). All three impaired word recall, a declarative task, but midazolam did so far more severely than the other two. All three also impaired procedural learning on a tracking task, and again midazolam had the strongest effect. The researchers noted that the modulating effects of these neurotransmitter systems on declarative and procedural memory were less system-specific than neuropsychological studies in brain-damaged patients had suggested.18PubMed. Dopamine-antagonistic, anticholinergic, and GABAergic effects on declarative and procedural memory functions In other words, while the brain structures involved in the two memory systems are anatomically distinct, the chemical messengers that modulate them overlap more than you might expect. A drug that changes GABA activity, for example, does not just hit one system — it blurs both, though to varying degrees.
Measuring Procedural Learning Is Harder Than You Think
One of the standard laboratory tools for measuring procedural learning is the serial reaction time task, in which people press buttons in response to stimuli that follow a hidden repeating pattern. Over time, people speed up on the repeating pattern compared to random sequences, which is taken as evidence that they have implicitly learned the pattern. The task has been used in hundreds of studies. But its reliability has come under scrutiny.
A meta-analysis of test-retest reliability found that the procedural learning effect measured by this task had a retest correlation well below acceptable psychometric standards (below 0.40), meaning that a person who shows a strong learning effect today might not show a similarly strong effect when retested.19PubMed Central. The reliability of the serial reaction time task: meta-analysis of test–retest correlations The authors describe this as a “reliability paradox” — the task reliably produces a group-level learning effect, but it does not reliably rank individuals. Within a single session, consistency was better (around 0.66), but that still leaves a lot of room for noise.
There is an even more fundamental concern. A separate study found that the speed improvement produced by the repeating sequence disappeared within minutes of training ending. The researchers concluded that the task may induce rapid temporary adaptation rather than genuine motor learning, since the sequence did not seem to be encoded in lasting memory.20PubMed. The “implicit” serial reaction time task induces rapid and temporary adaptation rather than implicit motor learning If the most widely used laboratory measure of procedural learning turns out to measure something transient rather than durable, that is a problem for the entire research literature that relies on it. The science here is still being worked out, and it is a reminder that measuring “knowing how” is inherently trickier than measuring “knowing that” — you cannot just ask someone to write down what they know.
The Distinction in Artificial Intelligence
The declarative-procedural split did not originate in psychology. It has deep roots in computer science and artificial intelligence, where the question of how to represent knowledge in a machine has been debated since the field’s early days. The debate crystallized into two camps. Proceduralists argued that knowledge is best embedded directly in programs — the system “knows” something by having a procedure that uses it. Declarativists argued that knowledge should be stored as a set of explicit facts separate from the procedures that manipulate those facts, with general-purpose reasoning engines operating over the fact base.21ScienceDirect. Representation and Understanding
This tension never fully resolved, and it persists in modern AI systems. Large language models, for instance, appear to store enormous amounts of declarative knowledge (facts, definitions, relationships) in their parameters, but their ability to carry out multi-step procedures reliably is notoriously fragile. Game AI faces the problem from the opposite direction: virtual characters need to execute complex action sequences (procedural) while making decisions informed by world knowledge (declarative). Bridging these two representations is a persistent engineering challenge, with tools specifically designed to translate between declarative planning domains and the procedural frameworks used by game engines.22Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. Bowyer: A Planning Tool for Bridging the Gap between Declarative and Procedural Domains The fact that AI researchers independently arrived at the same fundamental distinction that psychologists did suggests it reflects something deep about the structure of knowledge itself, not just the quirks of human brains.
Do Animals Have Declarative Knowledge?
No one doubts that animals have procedural knowledge. A rat that has learned to navigate a maze, a bird that has mastered a complex song, a dog that can catch a frisbee — these are all examples of procedural skill. The harder question is whether animals have anything resembling declarative knowledge: the ability to represent facts about the world or recall specific past events.
The evidence is mixed but increasingly suggestive. Research across a range of species, including primates, dolphins, scrub jays, rats, and pigeons, has found that some animals can recover memories of what happened, where it happened, and when it happened, meeting earlier criteria for episodic-like memory. Some can also answer unexpected questions about past events, which is another test designed to distinguish genuine memory retrieval from simple habit.23PubMed Central. Animals represent the past and the future Western scrub jays are the classic example: they cache food and later retrieve it in ways that suggest they remember what they stored, where they stored it, and how long ago. Whether this constitutes “declarative knowledge” in the human sense, or something functionally similar but experientially different, remains one of the more fascinating open questions in comparative cognition. The answer will depend partly on how tightly we want to tie the concept of “knowing that” to the ability to verbalize what you know.