The Tolman Cognitive Maps Experiment: Rats Know More Than They Show

The standard behaviorist account of how a rat learns a maze was clean and mechanical: the animal tries various routes, is rewarded at the end of one, and gradually strengthens the specific muscle movements and stimulus-response associations that led to the reward. Learning, on this view, is not mental — it is the stamping in of habits through reinforced repetition. Edward Tolman spent more than two decades running rats through mazes at UC Berkeley, and by 1948 he had accumulated enough evidence to argue that the behaviorist account was simply wrong. Rats, he showed, build mental models of their environment — cognitive maps — and use those maps flexibly when conditions change.

The Behaviorist Orthodoxy Tolman Was Challenging

In the 1920s and 1930s, American psychology was dominated by behaviorism, a school of thought that held that psychology should concern itself only with observable behavior, not with internal mental states. The leading figures — John Watson and B. F. Skinner — argued that learning could be fully explained without reference to anything happening "inside" the organism. Stimulus in, response out; reward strengthens the connection, punishment weakens it. The mind, if it existed at all, was irrelevant to a scientific account of behavior.

This was not merely a methodological preference. It was a philosophical position: mental states were either non-existent or causally inert epiphenomena. A rat running a maze was not "thinking about" the maze — it was executing stimulus-response chains. The route it took was not "remembered" — it was the strongest available habit, shaped by reinforcement history.

Stimulus-response (S-R) theory — the behaviorist model of learning, in which an organism learns by associating a specific stimulus with a specific response, with reinforcement strengthening the association. On this view, the learned item is a behavior, not a representation.

Tolman accepted the need for objective, observable evidence. He was as committed to empirical rigor as any behaviorist. But he rejected the philosophical restriction to behavior alone. He called his approach "purposive behaviorism" — behavior was still the data, but it pointed toward internal cognitive states that needed to be explained.

Source: Tolman, E. C. (1932), Purposive Behavior in Animals and Men (Appleton-Century-Crofts) — Tolman's early theoretical statement, arguing that behavior is organized around goals rather than S-R chains.

The Place Learning vs. Response Learning Controversy

One of the clearest empirical disputes between Tolman and the behaviorist framework centered on place learning versus response learning.

Response learning — learning a specific sequence of movements or turns (turn right, turn left, go straight) that leads to a reward. On the behaviorist account, this is how maze learning works: the animal learns the motor routine.

Place learning — learning the location of a goal in space, independent of the route used to reach it. On Tolman's account, this is what rats actually do: they learn where the food is, not which movements to make.

These two accounts make different predictions when you change the maze. If a rat has learned by response — "turn right, then left, then right" — it should be confused when the maze is changed so that the same sequence of turns no longer leads to food. If a rat has learned by place — "food is at the north end of the room" — it should be able to find the food by a new route if the old one is blocked.

Tolman conducted a classic test of this using a "cross maze" — a four-arm maze arranged like a plus sign (+), set in a room with distinctive visual cues on the walls. Rats were trained with food in the east arm, entering always from the south. A response-learner would learn "turn left" (which, from the south, leads east). A place-learner would learn "food is in the east arm."

When Tolman changed the entry point to the north arm, response-learners should still turn left — which now leads west, away from the food. Place-learners should turn right from the north entrance, which still leads to the east arm where the food was.

The rats predominantly turned toward the east arm — place learning, not response learning.

Source: Tolman, E. C., Ritchie, B. F. & Kalish, D. (1946), "Studies in spatial learning. I. Orientation and the short-cut," Journal of Experimental Psychology, 36(1), 13–24 — the place learning vs. response learning experiment using the cross maze.

Think of the difference this way: a response-learner navigating a familiar city block is memorizing a script — "walk forward, turn at the red door, turn again at the lamppost." A place-learner has a mental map and can detour around construction, take a shortcut through a new alley, or reach the destination from a different starting point. Tolman was arguing that rats were place-learners, not script-followers.

Latent Learning: The Unrewarded Rat Who Already Knows

The most influential series of Tolman's experiments demonstrated latent learning — learning that occurs without apparent reward and is not expressed in behavior until there is a reason to express it.

Latent learning — the acquisition of knowledge or representations without immediate reinforcement or behavioral expression. The knowledge exists in a latent state and becomes apparent only when it becomes relevant.

The experimental design was simple. Three groups of rats were placed in a complex maze.

• Group 1 (rewarded throughout): These rats received food at the goal box at the end of every trial. They learned quickly, reducing their errors consistently over the training trials.
• Group 2 (never rewarded): These rats received no food at the goal box. They were allowed to explore the maze but had no incentive to reach the goal. Their error rate was high and did not decrease consistently.
• Group 3 (reward introduced on day 11): These rats were in the maze for 10 days with no food reward, then on day 11, food was placed in the goal box.

The result was the key observation: on day 12 — after a single day of food reward — Group 3's error rate dropped sharply to match Group 1, which had been rewarded throughout. In fact, by some measures, Group 3 performed slightly better than Group 1.

If learning were purely the stamping-in of rewarded responses, Group 3 rats should have had to gradually learn the maze from scratch once food was introduced — they had received no reinforcement for successful navigation. Instead, they navigated as efficiently as rats that had been rewarded from the start, as if they had already learned the maze layout. They had — latently. The unrewarded exploration had built a cognitive map of the maze, and when a reason to use it appeared (food), the rats could apply it immediately.

Source: Tolman, E. C. & Honzik, C. H. (1930), "Introduction and removal of reward, and maze performance in rats," University of California Publications in Psychology, 4(17), 257–275 — the foundational latent learning experiment.

Source: Tolman, E. C. (1948), "Cognitive maps in rats and men," Psychological Review, 55(4), 189–208 — the synthesis paper that named the cognitive map concept and drew together the experimental evidence.

The Sunburst Maze: Shortcuts and Flexibility

A second major experiment, the "sunburst maze" or blocked-path experiment, tested whether rats could find shortcuts to a goal location — behavior that implies a flexible mental representation rather than a learned route.

Rats were trained to run a complex maze with multiple segments to reach a food reward. Then Tolman blocked the familiar route at a critical point and gave the rats access to a large array of new paths radiating outward from the blocked point, like the spokes of a wheel. Most paths led away from the goal location. One path led toward the general location of the goal.

Rats that had learned the maze flexibly — with a place-based cognitive map — should head toward the goal's location, even on a novel path. Rats that had learned only a specific motor sequence should be at a loss when the sequence is disrupted.

The rats predominantly chose the path pointing toward where the food had been. They were not navigating by remembered turns; they were navigating by location.

Source: Tolman, E. C., Ritchie, B. F. & Kalish, D. (1946), "Studies in spatial learning. II. Place learning versus response learning," Journal of Experimental Psychology, 36(3), 221–229 — the shortcut and blocked-path studies, second paper in the series.

This result is striking because the paths the rats chose were entirely novel — they had never been rewarded for going down them. Under a pure stimulus-response account, novel paths with no reinforcement history should not be chosen systematically. Under the cognitive map account, the rat picks the path that heads toward the map-location of the food, regardless of whether that specific path was ever rewarded.

The 1948 Paper: "Cognitive Maps in Rats and Men"

In 1948, Tolman synthesized the preceding two decades of maze research in a single paper — his presidential address to the American Psychological Association, published in Psychological Review as "Cognitive maps in rats and men." This paper named and defined the central concept, proposed the map analogy, and drew explicit implications for human psychology and social behavior.

Tolman described two types of cognitive maps, which he called "narrow strip maps" and "broad comprehensive maps." A narrow strip map is like the behaviorist's S-R chain — a specific route from A to B, with no information about the broader spatial context. A broad comprehensive map includes information about relative locations, distances, and relationships that allows flexible navigation from any starting point.

He argued that whether an organism develops a narrow or broad map depends on experimental conditions: stress, severe reward or punishment, and forced exposure to single routes favor narrow maps. Relaxed exploration of a rich environment favors broad maps. This was not just a claim about rats — Tolman ended the paper with an extended analogy to human social psychology, arguing that "narrow maps" in humans — oversimplified, rigid representations of the social world — underlie prejudice, aggression, and in-group/out-group hostility.

Source: Tolman, E. C. (1948), "Cognitive maps in rats and men," Psychological Review, 55(4), 189–208 — the full paper, which explicitly connects rat maze behavior to human cognitive and social phenomena.

This conceptual move — using animal learning research to make claims about human cognition — was itself a challenge to behaviorism, which had generally kept animal and human psychology strictly parallel (both governed by S-R laws) rather than allowing mental concepts to migrate from one to the other.

What Tolman's Theory Predicted That Behaviorism Could Not Explain

The central scientific advantage of the cognitive map hypothesis over S-R theory was predictive: the cognitive map account explained data that S-R theory could not.

Latent learning: S-R theory predicted that unrewarded maze exploration would produce no learning, because no response was being reinforced. Tolman's data showed that unrewarded rats learned the maze layout and applied it as soon as reward was introduced. S-R theorists spent years producing alternative explanations — arguing that latent exploration provided sub-threshold reinforcement, or that proprioceptive feedback from movement served as a reward — but none of these rescues were clean.

Place learning over response learning: S-R theory predicted that response learning (specific motor sequences) should be the dominant mode, since motor responses are what gets reinforced. Tolman's cross-maze and sunburst experiments showed place learning was predominant.

Shortcut taking: S-R theory predicted that novel routes with no reinforcement history would not be chosen systematically. Tolman's rats took shortcuts pointing toward goal locations they had never reached by that path.

Blocked-path detours: If a memorized route is blocked, an S-R learner has no information to guide alternative route selection. A cognitive map user has spatial information that points toward the goal location even from new starting points.

Source: Mackintosh, N. J. (1974), The Psychology of Animal Learning (Academic Press) — comprehensive review of the place learning vs. S-R learning debate and its resolution in the cognitive map framework.

The Neuroscience Vindication: Place Cells and the Hippocampus

Tolman proposed the cognitive map as a theoretical construct — he had no idea what the neural implementation looked like. The neural vindication came 23 years after his 1948 paper.

In 1971, John O'Keefe and Jonathan Dostrovsky, working at University College London, recorded electrical activity from individual neurons in the hippocampus (a brain structure in the medial temporal lobe) of freely moving rats. They found cells that fired specifically when the rat was in a particular location in its environment — regardless of which direction the rat was facing or which behavior it was performing. O'Keefe called these neurons place cells.

Place cell — a neuron in the hippocampus that increases its firing rate specifically when an animal is in a particular location in its environment. Different place cells respond to different locations; together, the population of active place cells encodes the animal's current position.

A collection of place cells covering an environment creates a neural map — exactly the kind of internal spatial representation Tolman had hypothesized. The hippocampus, it turned out, is the anatomical substrate of the cognitive map.

Source: O'Keefe, J. & Dostrovsky, J. (1971), "The hippocampus as a spatial map: preliminary evidence from unit activity in the freely-moving rat," Brain Research, 34(1), 171–175 — the discovery of place cells, the neural foundation of Tolman's cognitive map.

O'Keefe and Lynn Nadel expanded this into a full theory in their 1978 book The Hippocampus as a Cognitive Map, directly invoking Tolman's framework. In 2014, O'Keefe shared the Nobel Prize in Physiology or Medicine with May-Britt Moser and Edvard Moser, who had discovered grid cells — neurons in the entorhinal cortex that create a metric coordinate system for the cognitive map. The 2014 Nobel announcement explicitly cited Tolman's 1948 paper as the conceptual precursor.

Source: O'Keefe, J. & Nadel, L. (1978), The Hippocampus as a Cognitive Map (Oxford University Press) — the full neuroscientific theory, directly extending Tolman's framework.

Source: Nobel Assembly at Karolinska Institutet (2014), Nobel Prize in Physiology or Medicine press release — citing Tolman as the conceptual origin of the spatial mapping research rewarded in 2014.

Timeline of the Cognitive Map Research

Year	Event
1922	Tolman begins maze research at UC Berkeley
1930	Tolman & Honzik publish latent learning study — unrewarded rats learn maze layout
1932	Purposive Behavior in Animals and Men — Tolman's theoretical framework published
1946	Place learning vs. response learning studies published (Journal of Experimental Psychology)
1948	"Cognitive maps in rats and men" published in Psychological Review — the defining paper
1959	Tolman dies at age 72
1971	O'Keefe and Dostrovsky discover place cells in rat hippocampus
1978	O'Keefe and Nadel publish The Hippocampus as a Cognitive Map
1992	Grid cells discovered in entorhinal cortex by May-Britt Moser and Edvard Moser
2005	Moser lab publishes full description of grid cells in Nature
2014	Nobel Prize in Physiology or Medicine awarded to O'Keefe, May-Britt Moser, Edvard Moser — explicitly validating Tolman's framework

Relationship to Skinner's Behaviorism

Tolman's cognitive map work is the direct intellectual counterpart to B. F. Skinner's operant conditioning research, which was developing simultaneously and reached its most systematic form in the Skinner box experiments of the 1930s onward.

The contrast is instructive. Skinner's operant conditioning framework (see the Skinner box experiment) demonstrated that behavior is shaped by its consequences — reinforced behaviors increase in frequency, punished or unreinforced behaviors decrease. This is an extraordinarily powerful principle with genuine predictive force, and Skinner used it to account for an enormous range of animal and human behavior.

Tolman did not dispute that reinforcement affects behavior. He disputed that reinforcement was what was being learned. On Skinner's account, reward stamps in stimulus-response connections; the organism learns to do something. On Tolman's account, reward provides information about what is worth seeking; the organism learns something about the world and then decides what to do based on that knowledge.

The difference becomes visible exactly when you change conditions. Tolman's latent learning rats and place-learning experiments exposed the limits of a purely S-R account by showing that what had been learned was more flexible than any specific response chain. The rats navigated to goal locations, not by executing learned routines.

Modern psychology recognizes that both frameworks capture something real. Declarative memory — memory for facts and spatial layouts — follows something like the cognitive map model, with the hippocampus playing a central role. Procedural memory — memory for skills and habits — follows something more like the S-R model, mediated by different brain structures including the basal ganglia. Tolman and Skinner were, in a sense, each studying a different memory system.

Source: Squire, L. R. (2004), "Memory systems of the brain: a brief history and current perspective," Neurobiology of Learning and Memory, 82(3), 171–177 — review of multiple memory systems, contextualizing Tolman's and Skinner's frameworks within modern neuroscience.

What People Get Wrong About Tolman's Experiments

"Tolman rejected all of behaviorism." Tolman was trained as a behaviorist and maintained a commitment to objective, behavioral evidence throughout his career. He did not reject the importance of observable behavior; he rejected the philosophical claim that behavior was exhaustively explained by S-R associations. He called himself a "purposive behaviorist."

"Latent learning means reinforcement is unnecessary." Tolman's point was not that reinforcement is irrelevant to behavior — clearly it matters, as his rewarded Group 1 rats showed. His point was that learning (acquiring information about the world) occurs independently of reinforcement, and that reinforcement affects performance (whether learned information gets expressed in behavior) more than it affects learning per se.

"Cognitive maps are just memories of routes." A cognitive map in Tolman's sense is more than a recorded sequence of movements. It is a spatial representation that encodes the layout of an environment in a way that supports flexible navigation — including novel routes, shortcuts, and detours. A route memory is more like a strip map; a cognitive map is more like an aerial photograph.

"The hippocampus is only for cognitive maps." Modern neuroscience shows the hippocampus supports multiple forms of memory, including episodic memory (memory for specific past events) and some forms of associative learning. Spatial navigation is one domain in which hippocampal function is clear, but the structure is not limited to maps. The relationship between hippocampal function and Tolman's original concept is real but not exhaustive.

FAQs

Q: What was Tolman's main argument in the 1948 paper?
A: That rats navigating mazes learn flexible internal representations of their environment — cognitive maps — rather than chains of specific stimulus-response movements. The paper argued that animals form something like a mental model of spatial layout, which they can use to navigate flexibly, take shortcuts, and reach goals by novel routes.

Q: What is latent learning?
A: Learning that occurs without immediate reinforcement and is not expressed in behavior until there is a reason to do so. In Tolman's experiments, rats that explored a maze without food reward for 10 days could navigate it as well as always-rewarded rats on the very first day food was provided — suggesting they had learned the maze layout during unrewarded exploration.

Q: How does the cognitive map hypothesis differ from behaviorist S-R theory?
A: S-R theory holds that learning is the strengthening of specific stimulus-response associations through reinforcement — the organism learns to perform specific responses. The cognitive map hypothesis holds that learning can involve acquiring spatial or environmental representations without specific responses being reinforced — the organism learns about the world, then decides what to do. The difference becomes visible when conditions change and the organism must navigate in a new way.

Q: Did neuroscience confirm Tolman's cognitive map hypothesis?
A: Yes. In 1971, O'Keefe and Dostrovsky discovered place cells in the rat hippocampus — neurons that fire specifically when the animal is in a particular location, creating a neural map of the environment. Grid cells in the entorhinal cortex, discovered later, add a coordinate system to the map. These findings provided the neural substrate for exactly the kind of spatial representation Tolman had hypothesized. The 2014 Nobel Prize in Physiology or Medicine was awarded for this work.

Q: Did Tolman ever win a Nobel Prize?
A: No. Tolman died in 1959, before the neural basis of his cognitive map concept was established. The Nobel Prize for the place cell and grid cell work was awarded in 2014, 55 years after Tolman's death. Nobel Prizes are not awarded posthumously.

Q: How does Tolman's work relate to Skinner's operant conditioning?
A: They represent two competing accounts of what learning involves. Skinner argued that reinforcement shapes specific behaviors through stimulus-response associations. Tolman argued that organisms learn world-representations, not just behaviors. Modern neuroscience recognizes that both capture real memory systems: the hippocampal system supports flexible, map-like learning (closer to Tolman), while the basal ganglia support habit and procedural learning (closer to Skinner). Both systems operate in the same brain.

Q: What is a "narrow strip map" vs. a "broad comprehensive map"?
A: Tolman's terms for two types of cognitive representation. A narrow strip map encodes only a specific route from one point to another — like a set of turn-by-turn directions. A broad comprehensive map encodes the general layout of an environment and the spatial relationships among locations — like an aerial photograph. Tolman argued that stress, severe punishment, and forced single-route experience favor narrow maps; relaxed exploration favors broad ones.

The Takeaway

Tolman spent two decades watching rats navigate mazes under carefully controlled conditions, and what he showed was that the rats knew more than they were demonstrating. They mapped their environments, stored that map without needing a reward to do so, and applied it flexibly when conditions changed — exactly what a rigid chain of S-R associations cannot do. The cognitive map concept, proposed in 1948, waited 23 years for neuroscience to find the place cells and grid cells that implement it in the hippocampus, and then another 36 years for those discoveries to receive a Nobel Prize. Tolman did not live to see the vindication, but the rats had been right all along.

---

Read next:

• The Skinner Box Experiment
• The Pavlov Classical Conditioning Experiment
• The Köhler Chimpanzee Insight Experiment