Operant vs. Classical Conditioning

Operant vs. classical conditioning: the short answer

In classical conditioning, two stimuli are paired and a reflex transfers from one to the other. Pavlov's dogs already salivated at food; after a metronome repeatedly preceded food, they salivated at the metronome. Nothing the dog did changed what happened — food came regardless. In operant conditioning, a behavior is followed by a consequence and the behavior changes as a result. Skinner's rats pressed a lever, food arrived because they pressed, and pressing increased. The dog learned "metronome means food." The rat learned "pressing gets food." That difference — signal versus consequence, elicited reflex versus emitted action — is the whole distinction, and everything else in this article is a consequence of it.^[3]

The difference between classical and operant conditioning, side by side

How to tell which is which: a decision checklist

1. Name the response. Is it something the organism does with its skeletal muscles — walks, presses, says, buys — or something that happens to it — salivates, flinches, feels sick, feels afraid, heart races? Voluntary points to operant; reflexive or emotional points to classical.
2. Find the key event and check its timing. Does the important stimulus come before the response as a signal, or after it as a result? Before is classical; after is operant.
3. Test contingency. Would the outcome have happened anyway? If food comes whether or not the dog salivates, it is classical. If food comes only if the dog sits, it is operant.
4. State what was learned in one sentence. "X predicts Y" is classical. "Doing X produces Y" is operant.
5. Look for both. Most real scenarios contain a classically conditioned emotion and an operant action. Name each part separately rather than forcing one label on the whole scene.

Worked examples: classical or operant?

More practice: the examples page has more than fifty operant scenarios sorted by quadrant, and the quiz mixes classical and operant items with instant explanations.

Where people mix them up

• Deciding by whether the stimulus is pleasant. Both kinds of conditioning use pleasant and unpleasant stimuli. Decide by timing and contingency, not by valence.
• Calling the bell the unconditioned stimulus. The US is the stimulus that works without learning — the food. The bell (or metronome) is neutral, then conditioned.
• Assuming the CR is a copy of the UR. Often it is weaker; sometimes it is the opposite (Siegel's compensatory responses), and the CR to a shock-predicting tone is freezing, not the jump the shock itself produces.
• Filing negative reinforcement under classical conditioning because it "involves something unpleasant." Negative reinforcement is operant: a behavior removes an aversive stimulus and increases. Classical conditioning does not have reinforcement or punishment in Skinner's sense at all.
• Confusing the CS with the discriminative stimulus. A CS elicits a reflex regardless of behavior; an S^D signals that a behavior will be reinforced.
• Treating classical conditioning as mere pairing. Since Rescorla, the CS must predict the US. Pairings without predictive value produce little learning.
• Treating extinction as forgetting. In both kinds of conditioning, extinction is new learning that inhibits the old, which is why spontaneous recovery and renewal occur.
• Forcing one label on a scene that has both. "The dog gets excited at the leash and then sits for a treat" contains a classical part and an operant part. Full credit requires naming each.

For the people, dates, and disputes behind both traditions, see the history of operant conditioning.

Going further

The sections below go past the short answer: how classical conditioning really works, where the two kinds of learning meet, and the cases that blur the line.

Classical conditioning, explained properly

Pavlov's discovery

Ivan Pavlov was a physiologist studying digestion — work that earned him the 1904 Nobel Prize — when he noticed that his dogs began salivating before food arrived: at the sight of the food dish, at the footsteps of the attendant. He called these "psychic secretions" and spent the rest of his career studying them with the rigor of a physiologist, using a surgically implanted tube to measure drops of saliva. In many of his experiments the signal was a metronome, a buzzer, a light, or a touch rather than the bell of legend. The results were published in English in 1927 as Conditioned Reflexes.^[1] (Pavlov's own word was "conditional" — the reflex was conditional on the pairing — and "conditioned" is an early translation that stuck.)

The four terms

• Unconditioned stimulus (US or UCS): a stimulus that elicits a response without any learning. Food in the mouth.
• Unconditioned response (UR or UCR): the unlearned response to it. Salivation to food.
• Conditioned stimulus (CS): a previously neutral stimulus that, after predicting the US, elicits a response on its own. The metronome.
• Conditioned response (CR): the learned response to the CS. Salivation to the metronome — usually similar to the UR, but often weaker and sometimes different in form.

What Pavlov found

Acquisition is gradual: the CR grows over pairings. Extinction follows when the CS is presented repeatedly without the US — but Pavlov noticed that an extinguished response reappears after a rest (spontaneous recovery), which told him extinction was new learning laid over the old, not erasure. Modern work confirms this: extinguished responses also return when the context changes (renewal) or when the US is encountered again (reinstatement).^[4] A CR trained to one tone appears, more weakly, to similar tones (generalization); pairing one tone with food and another with nothing narrows the response to the first (discrimination). When Pavlov's laboratory made a circle-versus-ellipse discrimination progressively harder, a previously calm dog became agitated and uncooperative — the first "experimental neurosis."^[1] Finally, an established CS can itself condition a new stimulus (higher-order conditioning), which is how a word like "dinner" ends up doing what the metronome did.

Little Albert: the famous study and its problems

In 1920 John B. Watson and Rosalie Rayner reported conditioning fear in an infant, "Albert B.," about eleven months old. Albert initially reached for a white rat without fear. Watson then struck a steel bar with a hammer behind Albert's head whenever the rat appeared. After a handful of pairings, Albert cried and turned away at the sight of the rat alone, and his distress generalized to a rabbit, a dog, a fur coat, and a Santa Claus mask.^[5]

The study is in every textbook, and it should be read with its problems attached. Ethically, it deliberately induced fear in an infant who could not consent, and Watson and Rayner made no attempt to remove the fear before Albert left the hospital, though they knew in advance when he would leave. Methodologically, it was a single case with no control condition; fear was rated subjectively; some of Albert's reactions were mild or inconsistent; and the responses were "freshened up" with additional pairings between tests. A 1979 review found that textbooks had for decades embellished the results, reporting deconditioning that never happened and generalization that was never tested.^[6] Albert's real identity has been the subject of competing claims by historians. What survives is the modest, real finding: a fear response can be conditioned to a neutral stimulus in a human infant, and it generalizes.

It's not what you think it is: contingency, not pairing

The textbook story — "pair two stimuli enough times and the reflex transfers" — turns out to be wrong in an important way. In 1968 Robert Rescorla gave rats a tone followed by shock, but for some groups he added shocks during the silent periods too, so that the tone no longer predicted any change in the likelihood of shock. Those rats received exactly as many tone–shock pairings as the others, and they learned almost nothing.^[7] Leon Kamin's blocking experiment made the same point from another direction: if a light already predicts shock, adding a tone alongside it teaches the animal nothing about the tone, because the shock is no longer surprising.^[8] Rescorla and Allan Wagner turned these findings into a mathematical model in which learning is driven by prediction error — how much the outcome differs from what was expected.^[9]

Rescorla summarized the modern view in a 1988 paper whose title says it all: "Pavlovian conditioning: It's not what you think it is." Classical conditioning is not the mechanical transfer of a reflex by contiguity; it is the organism learning the predictive structure of its environment — which events signal which others — and adjusting a whole set of responses accordingly.^[10]

Taste aversion and biological preparedness

The other crack in the simple story came from John Garcia. In 1966 Garcia and Robert Koelling let rats drink "bright, noisy" water — sweetened, and accompanied by a light and a click with every lick. Rats that were then made ill (with X-rays or lithium chloride) later avoided the sweet taste but drank the noisy, bright water happily. Rats that were shocked instead avoided the light and click but not the taste.^[11] The animals were not equally ready to associate any stimulus with any outcome: tastes go with illness, sights and sounds go with pain. Taste aversion also broke the timing rule — it formed after a single trial with delays of an hour or more between taste and illness. Martin Seligman called this preparedness: evolution has made some associations easy to learn and others nearly impossible.^[12]

Operant conditioning, briefly

Edward Thorndike's cats, escaping from puzzle boxes, showed that responses followed by satisfying outcomes are "stamped in" — the law of effect.^[13] B. F. Skinner made this a laboratory science: an animal in a chamber, a lever or key, a consequence delivered by the apparatus, and a cumulative record of responses over time. In a 1937 paper he formally separated the two kinds of learning, calling Pavlov's "Type S" (stimulus-elicited, respondent) and his own "Type R" (response-emitted, operant).^[3] Every consequence falls into one of four quadrants — positive reinforcement, negative reinforcement, positive punishment, negative punishment — and how often it arrives is governed by schedules of reinforcement. Behavior that no longer pays off undergoes extinction, complex behavior is built by shaping, and the antecedent that signals when a behavior will be reinforced brings it under stimulus control. The complete guide to operant conditioning ›

How classical and operant conditioning work together

Outside the laboratory the two processes almost never run alone, and the most useful accounts of real behavior combine them.

Two-factor theory of avoidance

Why does a rat keep jumping a barrier to avoid a shock that never comes any more? O. H. Mowrer's answer was two processes in sequence: first, a warning signal is classically conditioned to elicit fear; second, the avoidance response is operantly reinforced by escape from that fear.^[14] The same logic explains why phobias persist: the fear may be acquired classically (a dog bite, or even a frightening story), but it is maintained operantly, because avoiding dogs is negatively reinforced by relief and the fear never gets a chance to extinguish. Exposure therapy is the deliberate blocking of the operant half so the classical half can extinguish.

Conditioned reinforcers are classically conditioned

A clicker is silent to a dog that has never heard one paired with food. Pair click and treat a few dozen times and the click acquires two properties at once. It is a CS: it elicits the anticipatory excitement that food elicits. And it is a conditioned reinforcer: delivered right after a behavior, it strengthens that behavior, bridging the seconds until the treat arrives.^[15] Money, praise, grades, and the green checkmark work the same way — classically conditioned value, operantly deployed. How clicker training uses both ›

Conditioned suppression

One of the cleanest laboratory demonstrations of the two processes interacting is also one of the oldest. In 1941 Estes and Skinner had rats pressing a lever for food, then sounded a tone that ended in shock. As the tone acquired fear (classical), the rats' ongoing lever pressing slowed or stopped during it (operant behavior suppressed).^[16] The "conditioned emotional response" became a standard way to measure Pavlovian fear — by its effect on operant behavior.

Addiction cues

Drug taking is operant: the drug's effects reinforce the behavior that produces them. But the paraphernalia, the place, and the people become classically conditioned cues. Shepard Siegel showed that rats given morphine in a familiar environment developed conditioned compensatory responses — the body preparing to counteract the drug — so that tolerance was partly a learned response to the setting, and the same dose in a new place was more dangerous.^[17] Craving triggered by cues, and relapse in old environments, is classical conditioning driving the person back toward the operant.

Pavlovian-instrumental transfer

A classically conditioned cue can also change the vigor of operant behavior without ever having been part of it. Train a rat to press a lever for food; separately, in another session, pair a tone with free food. Then play the tone while the rat is pressing: pressing speeds up, even though the tone was never a signal for pressing and pressing has never paid off during it. William Estes reported the effect in 1948, and it is now called Pavlovian-instrumental transfer (PIT).^[18][19] It is the laboratory version of a familiar experience: the smell of the bakery does not teach you to walk in, but it makes you walk in faster. In addiction research PIT is one of the main models of how drug cues energize drug seeking.

Behavior without reinforcement? Autoshaping and contrafreeloading

Some observations look, at first, like operant behavior that no reinforcement produced, and they mark the edge of the law of effect.

Autoshaping and sign-tracking

In 1968 Brown and Jenkins lit a pigeon's response key for a few seconds and then delivered grain — whether or not the bird did anything. After a few dozen pairings the pigeons began pecking the lit key. Nobody had shaped the peck; the bird had "auto-shaped" it.^[20] The following year Williams and Williams arranged that pecking the key cancelled the grain, so that the only way to be fed was not to peck. The pigeons kept pecking — less, but persistently — and lost food for it.^[21] This omission result rules out reinforcement as the cause: the pecks were being punished by food loss and continued anyway. Jenkins and Moore then showed that the form of the peck matched the reinforcer — birds autoshaped with grain pecked the key as if eating it, birds autoshaped with water pecked as if drinking.^[22] The behavior is directed at the signal as if it were the reward, which is why Hearst and Jenkins named it sign-tracking.^[23]

The accepted interpretation is that autoshaping is classical conditioning: the key light is a CS, the grain a US, and approaching and pecking a food-predicting stimulus is the conditioned response, as inevitable in a pigeon as salivation in a dog. The autoshaping procedure has, in fact, become one of the standard ways to measure Pavlovian conditioning. It also uncovered stable individual differences: some rats become sign-trackers who approach the cue, others goal-trackers who go straight to the food cup, and dopamine appears to be required for the first kind of learning but not the second.^[24] The lesson for the operant–classical distinction is not that the law of effect is wrong but that a response can look operant and be Pavlovian, and the experimenter's job is to find out which contingency is actually controlling it.

Contrafreeloading

Give a rat free food in a dish and a lever that delivers the same food, and it will press the lever for a substantial share of its meals. Jensen reported the effect in 1963, and it has been found in most species tested, with cats the notable exception.^[25][26] Working for food that is freely available is not what a naive reading of reinforcement predicts. It is compatible with a fuller one: the opportunity to explore, manipulate, and gather information is itself reinforcing, and a well-designed environment for a captive animal — or a person — is one that lets it work.

Key takeaways

• Classical conditioning pairs two stimuli so that a reflex transfers from one to the other; operant conditioning follows a behavior with a consequence that changes its future rate. Signal versus consequence, elicited reflex versus emitted action, is the whole distinction.
• To tell them apart, name the response (reflexive or voluntary), check whether the key event comes before or after it, and test contingency. If the outcome would have happened anyway, it is classical; if it depends on the behavior, it is operant.
• A conditioned stimulus elicits a response regardless of behavior; a discriminative stimulus sets the occasion for a behavior whose outcome still depends on it. Negative reinforcement is operant, not classical, because a behavior removes the aversive stimulus and increases.
• Classical conditioning is prediction, not mere pairing: a stimulus that does not predict the outcome teaches almost nothing. Taste aversion shows that some associations are biologically prepared, forming in one trial across delays of an hour or more.
• The two processes usually work together. A phobia is acquired classically and maintained operantly by avoidance, a clicker is a conditioned stimulus and a conditioned reinforcer at once, and autoshaping shows that a response can look operant and be Pavlovian.

Check yourself

You buckle your seat belt to stop the chime. Because the chime is unpleasant, a classmate files this under classical conditioning. Is that right?

No. This is operant conditioning, specifically negative reinforcement: a voluntary behavior removes an aversive stimulus and becomes more frequent. Whether a stimulus is pleasant or unpleasant does not decide the type of learning; timing and contingency do, and classical conditioning has no reinforcement in Skinner's sense at all.

A pigeon pecks a lit key that is followed by grain no matter what the bird does. Since the peck looks like a lever press, is it operant behavior?

No. This is autoshaping, and the accepted interpretation is classical conditioning: the key light is a conditioned stimulus for grain, and pecking a food-predicting signal is the conditioned response. Pigeons keep pecking even when a peck cancels the grain, which rules out reinforcement as the cause.

A cat comes running when it hears the can opener. Classical or operant?

Both. The sound is a conditioned stimulus that elicits excitement and salivation, which is classical; running to the kitchen is an operant reinforced by food. Full credit means naming each part rather than forcing one label on the scene.

In Pavlov's experiment, which is the unconditioned stimulus: the metronome or the food?

The food. The unconditioned stimulus is the one that elicits the response without any learning, and food in the mouth produces salivation from the start. The metronome begins as a neutral stimulus and becomes a conditioned stimulus only after it reliably predicts the food.

Explain it to a friend. Explain the difference between the two kinds of conditioning using a single everyday example that contains both, and say which part is which without using the words voluntary or involuntary.