Higher-order conditioning is a form of learning in which a stimulus acquires emotional or motivational significance without ever being directly paired with a reward or punishment. Instead, it piggybacks on an earlier round of learning: an animal (or person) first learns that one cue predicts something important, and then a second, completely neutral cue gets linked to that already-conditioned first cue. The result is that the second cue starts triggering responses on its own, even though it was never connected to the original event. The phenomenon extends across a remarkably wide range of species and has real consequences for understanding everything from phobias to addiction, though reliably producing it in a laboratory with human participants has turned out to be surprisingly tricky.
How It Differs from Ordinary Conditioning
In standard Pavlovian conditioning, the recipe is straightforward: pair something neutral (a tone, a light, an image) with something that already matters biologically (food, a mild shock), and the neutral thing starts producing a response on its own. The key ingredient is direct contact with the motivationally significant event, what researchers call the unconditioned stimulus. Higher-order conditioning breaks that rule. It shows that stimuli can pick up motivational properties through indirect chains of association, without the learner ever experiencing them alongside the primary reinforcer.
The most studied version is second-order conditioning. In a typical experiment with rats, you first train the animal that a light predicts food. Once the rat reliably responds to the light, you introduce a tone that always appears just before the light, but no food is delivered during this phase. Despite never being paired with food, the tone begins to elicit food-anticipation behaviors. The light acts as a stand-in reinforcer, lending its learned value to the new cue.1PubMed. The basolateral complex of the amygdala is necessary for acquisition but not expression of CS motivational value in appetitive Pavlovian second-order conditioning This chaining effect is what makes higher-order conditioning so significant: it dramatically multiplies the number of cues in the environment that can trigger emotional or motivational reactions, all without any new direct experience with the thing that originally mattered.
The Two Main Procedures
Higher-order conditioning actually encompasses two distinct laboratory procedures that produce similar-looking outcomes through different training sequences. In second-order conditioning, the first-order association is established first (tone predicts food), and then a new stimulus is layered on top (light precedes the already-conditioned tone). In sensory preconditioning, the order flips: you first expose the animal to a pairing of two neutral stimuli (light and tone presented together when neither means anything yet), and only afterward do you condition one of them (tone now predicts food). When you later test the light, it also produces a food-related response, even though the light-tone pairing happened before either stimulus meant anything.2PubMed Central. Higher-Order Conditioning and Dopamine: Charting a Path Forward
Both procedures demonstrate that the brain builds representations of relationships between stimuli and can propagate learned value across those relationships. But they do not appear to work through identical mechanisms, which matters for understanding the underlying learning. Second-order conditioning and first-order Pavlovian conditioning are affected differently by the same experimental manipulations, and the conditioned responses that emerge from higher-order conditioning often do not mirror what you would see if the stimulus had been directly paired with the reinforcer itself.3PubMed Central. Higher-Order Conditioning: What Is Learnt and How it Is Expressed This is a subtle but important point: higher-order conditioning produces genuine behavioral change, but the nature of that change is not simply a diluted copy of first-order conditioning.
Why It Is So Hard to Show in Humans
If you have read about second-order conditioning in a textbook, the examples are almost certainly from animal studies. There is a good reason for that. While the phenomenon has been robustly demonstrated in rats, pigeons, honeybees, and other non-human animals, producing clean evidence of second-order conditioning in human participants has been remarkably difficult. A review of the existing human literature concluded that second-order conditioning is “a real but difficult phenomenon to obtain in humans.”4PubMed Central. Second-Order Conditioning in Humans
Several factors conspire to make human demonstrations elusive. One major culprit is conditioned inhibition, which is essentially the opposite of second-order conditioning and can arise from the same training procedure. Here is the problem: during the second phase of training, the new stimulus (call it X) appears alongside the already-conditioned stimulus (call it A), but the original reinforcer is absent. The intended lesson is “X predicts A, which predicts reward.” But humans, being pattern-detectors, may instead learn “when X appears with A, there is no reward, so X must signal the absence of reward.” That transforms X into an inhibitor rather than an exciter. Whether a given training setup produces second-order conditioning or conditioned inhibition depends heavily on the balance and timing of trials.
Research has shown this competition directly. Second-order conditioning is most likely to appear early in training, when the first-order stimulus has been heavily reinforced but the second-order pairing trials are still few. As training continues and the learner accumulates more trials where X appears alongside A without the reinforcer, conditioned inhibition takes over.5PubMed Central. Second-Order Conditioning and Conditioned Inhibition in Different Moments of the Same Training: The Effect of A+ and AX- Trial Number In other words, there is a window of opportunity. Run too few second-phase trials and you may not detect the effect; run too many and you accidentally train the participant to treat the second-order cue as a safety signal. Getting the parameters right is genuinely difficult, and many early human studies likely overshot the window without realizing it.
What Happens When the First-Order Cue Loses Its Power
A natural question is whether second-order conditioning survives if you extinguish the first-order response. If a rat learns that a light predicts food, then a tone predicts the light, what happens to the tone’s ability to trigger a response if you now repeatedly present the light without food until the rat stops responding to it? The answer turns out to depend on the conditions of extinction in interesting ways.
In one experiment, rats that were allowed to perform their normal consummatory behavior (licking water) during the extinction of the first-order stimulus showed reduced second-order responding afterward, as you might expect. But rats that were not allowed to engage in that behavior during first-order extinction maintained their second-order conditioned response even though the first-order response had been weakened.6Japanese Psychological Research. The effect of on‐ or off‐line extinction of a first‐order conditioned stimulus on a second‐order conditioned response in rats This suggests that second-order conditioning does not simply ride on the current strength of the first-order association. Once established, the second-order response can develop a degree of independence, a finding that has implications for understanding why fears and cravings can persist even after the original triggering cue has been defused through therapy or experience.
The Brain Circuitry Behind It
The amygdala, and specifically a region called the basolateral complex, plays a central role in higher-order conditioning, particularly for emotional and motivational learning. In appetitive (reward-based) second-order conditioning, the basolateral amygdala is needed when the animal is first acquiring the motivational value of the second-order cue. Rats with lesions to this area before the second-order phase failed to develop conditioned responses to the new stimulus, even though their first-order conditioning was intact.7PubMed. The basolateral complex of the amygdala is necessary for acquisition but not expression of CS motivational value in appetitive Pavlovian second-order conditioning
On the threat side, recent work has mapped out a network involving the amygdala and a region called the posterior piriform cortex. Blocking a particular type of receptor (NMDA receptors, which are important for memory formation) in the basolateral amygdala after second-order threat conditioning significantly reduced freezing to the second-order cue when tested the following day.8Communications Biology. Second-order threat conditioning in the amygdala-posterior piriform cortex network This tells us that the amygdala is not only needed during the learning phase but is actively involved in consolidating the memory of the second-order association. The brain treats these indirect threat memories as worth storing, which makes evolutionary sense: if a rustling sound reliably precedes the sight of a predator, it pays to fear the rustling even though the sound itself has never hurt you.
The role of dopamine in higher-order conditioning has also attracted attention. Dopamine neurons are famous for encoding prediction errors in first-order conditioning, firing when a reward is better or worse than expected. How these signals operate during higher-order learning, where the “reward” is itself just a learned cue, is an active area of investigation. Some findings challenge simple prediction-error models and suggest that the brain builds richer internal models of the relationships between stimuli rather than just tracking reward expectation.9PubMed Central. Higher-Order Conditioning With Simultaneous and Backward Conditioned Stimulus: Implications for Models of Pavlovian Conditioning
Implications for PTSD and Anxiety
Higher-order conditioning provides a compelling framework for understanding how anxiety disorders, and PTSD in particular, can spiral outward from an initial traumatic experience. In PTSD, trauma-relevant cues (a car backfiring, the smell of gasoline) can function as powerful conditioned stimuli. The critical insight from a higher-order conditioning perspective is that those trauma-associated cues can then act as reinforcers themselves, spreading conditioned fear to entirely new stimuli that were never present during the original trauma.
A study comparing people with PTSD to trauma-exposed controls found evidence consistent with this process. People with PTSD showed enhanced second-order conditioning and impaired extinction, meaning that new cues readily acquired fear associations through contact with trauma-related stimuli, and those associations were hard to undo.10PubMed. Failure of extinction of fear responses in posttraumatic stress disorder: evidence from second-order conditioning This helps explain one of the most frustrating features of PTSD: the number of situations that trigger distress tends to expand over time rather than shrink, even in the absence of new traumatic events. It is not that new traumas keep occurring; it is that existing trauma cues keep conditioning new ones.
This also has practical consequences for treatment. Standard exposure therapy works by repeatedly presenting the feared cue without the bad outcome, hoping to extinguish the conditioned response. But if the feared cue has already spawned a network of second-order (and potentially third-order) conditioned stimuli, extinguishing the original cue may not be enough. The downstream cues can maintain the fear network independently, as the extinction research described earlier suggests.
Addiction and Smoking Cues
A similar expansion of cue networks plays out in addiction. Higher-order conditioning helps explain why cravings can be triggered by stimuli that seem only distantly related to actual drug use. In a study with smokers and non-smokers, participants viewed abstract geometric figures that had been paired with either smoking-related images or neutral images. The figures paired with smoking imagery elicited larger brain responses in smokers, but not in non-smokers, even though the geometric shapes had never been directly associated with nicotine or the act of smoking.11PubMed Central. Electrophysiological correlates of associative learning in smokers: a higher-order conditioning experiment
This finding illustrates something anyone who has tried to quit smoking or another habit intuitively understands: the triggers multiply. It starts with the obvious cues like the smell of tobacco or seeing a cigarette. Through higher-order conditioning, those cues lend their motivational power to whatever else happens to be around, whether that is a particular time of day, a coffee cup, or a friend’s living room. The web of craving-triggering associations grows with experience, making relapse harder to prevent simply by avoiding the most obvious reminders.
Higher-Order Conditioning Across Species
One of the more remarkable aspects of higher-order conditioning is how far down the evolutionary tree it extends. It is not limited to mammals with complex brains. Honeybees, with brains containing roughly one million neurons, demonstrate second-order conditioning in carefully controlled experiments. After learning that one odor predicts sugar, bees transferred responding to a new odor that had been paired with the already-conditioned one. About 40% of bees responded to the second-order odor when it was presented in the correct forward sequence (new odor followed by conditioned odor), compared to only 20% when presented in the reverse order or paired with a non-reinforced odor.12PubMed Central. Forward and backward second-order Pavlovian conditioning in honeybees
The fact that forward pairing (new cue, then conditioned cue) worked but backward pairing was much less effective mirrors what is found in vertebrates and tells us something about the temporal logic of this kind of learning. The brain, even a very small insect brain, seems to interpret “X comes before A” as “X predicts A” while resisting the reverse inference. Bee research has also established that their mushroom bodies, brain structures involved in learning and memory, are necessary for more complex configural discriminations in higher-order tasks.13PubMed Central. Neural substrate for higher-order learning in an insect: Mushroom bodies are necessary for configural discriminations The wide distribution of this learning ability across such different nervous systems suggests it is an ancient and fundamental capacity rather than a sophisticated cognitive trick limited to big-brained species.
The Methodological Tangle
Studying higher-order conditioning has been dogged by a recurring methodological problem: it is hard to be sure that what you are seeing is genuinely higher-order conditioning rather than some other process producing a similar-looking result. Pseudoconditioning (where an animal becomes generally sensitized and responds to any stimulus) and mediated learning (where the animal mentally reactivates the first-order memory during second-order training, essentially imagining the reinforcer) can both mimic the behavioral output of true higher-order conditioning.14PubMed Central. Editorial: Higher-Order Conditioning: Beyond Classical Conditioning Ruling these alternatives out requires careful control conditions and has been a preoccupation of the field since its earliest days.
This methodological difficulty is part of why the human literature is so thin. With animal subjects, you can run hundreds of training trials and use physiological measures. With human participants, you are constrained by shorter sessions, more variable attention, and the fact that people tend to reason about what is happening in the experiment, which can either facilitate or interfere with the conditioning process in ways that are hard to control. The field is still working out what the critical parameters are for reliably producing the effect in humans, including how many first-order trials to run, how many second-order pairings are optimal before conditioned inhibition takes over, and what kinds of stimuli work best.
What Higher-Order Conditioning Reveals About How the Brain Learns
Beyond its applied relevance, higher-order conditioning has become an important testing ground for theories about how learning works at a computational level. Simple prediction-error models, which assume the brain updates its expectations based on the gap between what was predicted and what actually happened, have trouble accounting for some higher-order conditioning results. For instance, when the second-order stimulus is presented simultaneously with or even after the first-order stimulus (backward conditioning), some learning still occurs, which should not happen under strict prediction-error rules.15PubMed Central. Higher-Order Conditioning With Simultaneous and Backward Conditioned Stimulus: Implications for Models of Pavlovian Conditioning
Findings like these have pushed researchers toward models in which the brain builds rich internal representations of the task structure, not just keeping a running tally of prediction errors. The animal does not merely learn “this cue predicts reward”; it seems to learn something closer to “here is how these stimuli relate to each other in time and space, and here is where reward fits into that map.” This kind of model-based learning, as opposed to simple trial-and-error updating, would naturally explain why second-order conditioning can survive extinction of the first-order cue and why the temporal ordering of stimuli matters so much. The bee that learns “odor C precedes odor A, and odor A leads to sugar” appears to be representing a sequence, not just accumulating associative strength.
Research on amnesia provides another angle on this. In tasks that require learning complex sequential patterns, people with amnesia show deficits specifically in picking up the higher-order structure of the sequence, even when their overall learning scores look normal on the surface.16MIT Press Journals (Journal of Cognitive Neuroscience). Higher-order associative learning in amnesia: evidence from the serial reaction time task This suggests that the brain systems supporting higher-order learning are at least partly distinct from those handling simpler associations, and that damage to memory structures can selectively impair the ability to build the layered representations that higher-order conditioning depends on.
Everyday Life and Why This Matters Outside the Lab
Higher-order conditioning operates constantly in daily experience, though you rarely notice it. Consider how the jingle of an ice-cream truck acquires its pull. A child first learns that ice cream is delicious (first-order). Then the jingle becomes associated with seeing the truck (second-order). Eventually, just thinking about the neighborhood where the truck usually parks can start a craving (potentially third-order). Each layer is more removed from the original experience of eating ice cream, yet each can independently motivate behavior.
Brand marketing relies heavily on this principle, even when marketers do not use the term. A logo is repeatedly paired with pleasant imagery, music, or celebrity faces (first-order conditioning of the brand). Over time, the logo itself becomes a conditioned stimulus powerful enough to lend positive feelings to entirely new products the brand attaches itself to. The new product benefits from second-order conditioning: it was never directly paired with the pleasant imagery, but it inherits the warm glow through the brand logo.
The darker side of this process is visible in how trauma and addiction propagate through ever-widening networks of cues, as described in the clinical sections above. But even in mundane contexts, higher-order conditioning shapes preferences and aversions in ways people rarely trace back to their source. A persistent dislike of a particular style of music might stem not from anything about the music itself, but from its association with a person or place that was itself associated with an unpleasant experience. The chain of associations can be long enough that the connection to the original event is completely opaque, yet the emotional response remains strong. Recognizing that these chains exist, and that they can form without conscious awareness, is one of the more useful takeaways from a body of research that has been quietly accumulating in laboratories for over a century.