What Is the Difference Between Shaping and Chaining?

Shaping and chaining solve different teaching problems. Shaping builds a single behavior that does not yet exist by reinforcing closer and closer versions of it until the target form appears. Chaining links a series of already-learned behaviors into a reliable sequence, so the learner performs a multi-step task from start to finish. The confusion between them is understandable because both involve breaking something complex into smaller pieces and reinforcing progress. But the logic runs in opposite directions, and picking the wrong one for the job can stall learning or produce frustrating side effects.

How Shaping Works

Shaping starts with whatever the learner is already doing and gradually moves the goalposts. A speech therapist working with a nonverbal child might first reinforce any vocalization at all, then only vocalizations that sound roughly like a target word, then only clear approximations of that word. At each stage, the criterion tightens: what earned reinforcement yesterday no longer does today. The technique is sometimes called “successive approximations” because each reinforced version is a step closer to the final goal.

The key feature of shaping is that the target behavior does not have to exist before training begins. You are not teaching the learner to string together steps they already know. You are coaxing a new response into existence by selectively reinforcing natural variation in what they do. A rat does not “know” how to press a lever before shaping begins. It wanders near the lever, and that gets reinforced. Then it touches the lever, and that gets reinforced. Then only a full press counts. The behavior emerges through this progressive tightening of criteria.

Timing matters enormously. If you raise the criterion too fast, the learner stops getting reinforced and may give up or become frustrated. If you raise it too slowly, you can get stuck reinforcing a crude version of the behavior that becomes so strong it resists further refinement. Skilled shapers read the learner’s behavior in real time and adjust the pace accordingly, which is part of why shaping is sometimes described as an art as much as a science.

How Chaining Works

Chaining assumes that each individual step in a sequence already exists in the learner’s repertoire, or can be prompted into existence. The challenge is not producing a novel behavior but connecting multiple known behaviors so that completing one step becomes the cue for the next. Think of getting dressed: pulling a shirt over your head, pushing your arms through sleeves, and pulling the shirt down are all movements a child can perform. Chaining turns that collection of movements into a smooth, self-sustaining routine.

The mechanism that holds a chain together is conditioned reinforcement. Each step’s completion produces a stimulus, like the sight of your arm emerging from a sleeve, that has been paired with eventual reinforcement. That stimulus both signals “you’re on track” and motivates the learner to keep going. Research on how these intermediate signals function has shown that stimuli paired with reinforcement can serve as conditioned reinforcers even without being established as discriminative stimuli through other means, and that these middle-link cues can drive surprisingly robust responding.

1PubMed Central. A review of positive conditioned reinforcement

In pigeon experiments studying chain schedules, researchers found that birds sometimes responded more vigorously in the early links of a chain than in the final link closest to food, a pattern that only makes sense if the middle-link stimulus itself was acting as a powerful reinforcer.

2PubMed. Conditioned reinforcement in chain schedules when time to reinforcement is held constant

Forward, Backward, and Total Task Presentation

Chaining comes in three common flavors, and the differences matter for practical decisions about how to teach a skill.

  • Forward chaining: You teach step one first, then step one plus step two, then steps one through three, and so on. The learner always starts at the beginning and works forward. The trainer provides help with later steps the learner has not yet reached.
  • Backward chaining: You start by teaching the last step. The trainer completes everything up to the final step, and the learner does only that last piece, earning the reinforcer. Then the learner does the last two steps, then the last three, and so on. One advantage here is that the learner always finishes the task independently, which means every trial ends with the reinforcer following their own effort.
  • Total task presentation: The learner attempts the entire sequence on every trial from the start, with prompts and assistance as needed on steps they have not mastered. This is closer to how most people learn everyday skills in natural settings.

There is no universal winner among these three. A study comparing forward and backward chaining in children found that learners acquired target skills under both conditions, but no consistent advantage emerged for either direction across individual children.

3PubMed Central. An assessment of the efficiency of and child preference for forward and backward chaining

That said, total task presentation has shown an edge in some comparisons. A study training individuals with severe disabilities found that total task presentation was superior to backward chaining for rate of learning.

4Education and Training of the Mentally Retarded. Comparisons of Backward Chaining and Total Task Presentation in Training Severely Handicapped Persons

The practical takeaway is that the “best” chaining direction depends on the learner, the task, and sometimes on which version the learner prefers. Some children clearly enjoy backward chaining because they always get to be the one who finishes the job. Others find total task presentation more motivating because they get to practice the whole thing every time. When the evidence does not point clearly one way, learner preference is a reasonable tiebreaker.

When to Use Shaping and When to Use Chaining

The decision usually comes down to what is missing. If the learner cannot perform the target response at all, even a rough version of it, shaping is the tool. You cannot chain together steps that do not exist. But if the individual steps are already in the learner’s skill set and the problem is linking them into a smooth sequence, chaining is the better fit.

In practice, many real-world teaching programs use both. Teaching someone to cook an omelet might involve chaining the overall sequence (crack eggs, whisk, pour into pan, fold, plate) while using shaping within one or more steps. If the person has never cracked an egg before, you shape that component behavior first, then slot it into the chain. A chapter on teaching office skills to individuals with intellectual disabilities frames this combined approach explicitly: task analysis breaks the job into steps, chaining connects the steps, and shaping builds proficiency within each step by reinforcing successive approximations.

5IGI Global. Task Analysis, Chaining, and Shaping: Teaching Job-Ready Office Skills to Individuals With Intellectual Disability

A useful mental shortcut: if you can write a task analysis listing discrete steps, you are probably looking at a chaining problem. If what you need is a single behavior that just is not good enough yet, or does not exist at all, you are looking at a shaping problem.

Clinical Applications in Autism and Developmental Disabilities

Both techniques have decades of use in clinical settings, particularly in applied behavior analysis programs for children with autism. Their effectiveness depends heavily on how they are implemented, and some subtle differences in approach can produce very different outcomes.

For building vocal language in children who are mostly nonverbal, shaping has shown clear advantages when done carefully. A study of preschoolers with autism compared a standard shaping procedure, where the therapist systematically reinforced closer approximations of target sounds, against a looser “reasonable attempts” condition. The children mastered vocal skills faster under the shaping condition.

6PubMed Central. Is a reasonable attempt reasonable? Shaping versus reinforcing verbal attempts of preschoolers with autism

But the picture is not entirely straightforward. An earlier study examined the difference between reinforcing successive motor approximations of speech sounds versus simply reinforcing any attempt to produce speech. The results favored the less structured approach: children showed more interest, enthusiasm, and happiness during sessions, and their speech production actually improved more when attempts were reinforced broadly rather than shaped according to strict motor criteria.

7PubMed. Producing speech use in nonverbal autistic children by reinforcing attempts

These two findings are not as contradictory as they seem. The first study compared shaping against a specific “reasonable attempts” protocol, while the second compared motor-specific shaping against a broader motivational approach. The lesson is that how tightly you define your shaping criteria matters. Shaping that is too rigid about the motor form of a sound can undermine the child’s willingness to try. Shaping that gradually tightens criteria while keeping the child engaged tends to work well. The emotional state of the learner is not a side issue; it is part of whether the procedure works.

Chaining, meanwhile, dominates when the goal is teaching multi-step self-care and daily living skills. Dressing routines, tooth brushing sequences, meal preparation, and workplace tasks all lend themselves to chaining because they are inherently sequential. The therapist breaks the task into discrete steps through a task analysis, then teaches those steps in sequence using whichever chaining direction suits the learner.

Animal Training

Professional animal trainers rely on both techniques constantly, though they do not always use the textbook terminology. Shaping is fundamental to clicker training: the trainer clicks and rewards successive approximations of a target behavior, gradually raising criteria. Chaining is what turns individual tricks into routines, like a dog performing a sequence of sit, shake, spin, and bow on a single cue.

Backward chaining has found a particularly interesting niche in addressing behavior problems. A study on companion dogs with object-guarding aggression used clicker training with a backward-chaining procedure to teach an alternative behavior sequence. After implementation, success rates on the target response more than doubled in all four dogs, and aggressive responses dropped in frequency by roughly 58 to 69 percent across the group.

8Journal of Veterinary Behavior. Using clicker training as a treatment of object guarding in 4 companion dogs (Canis lupus familiaris)

The backward-chaining approach worked well there because the final step of the chain, voluntarily moving away from the guarded object and receiving a treat, was the most reinforcing part. By teaching that step first and building backward, each new step the dog learned led directly into an already-fluent, already-reinforced sequence. The dog was never stuck performing an unfamiliar first step with an uncertain outcome far down the line.

Common Mistakes and Failure Modes

Shaping goes wrong most often when the trainer moves too fast. If the criterion jumps from “any vocalization” to “a clear word” in one leap, the learner fails repeatedly, stops trying, and may develop avoidance of the entire training context. The opposite error, lingering too long on a lenient criterion, produces a different problem: the reinforced approximation becomes so strong that it resists change. A dog that has been heavily reinforced for touching a lever with its nose may struggle to learn to press it with a paw, because the nose-touch is now deeply ingrained.

Chaining has its own characteristic failure mode: superstitious behavior. When a learner performs a chain of responses, errors or extra behaviors can get accidentally woven into the sequence because they happen to occur before the next reinforced link. Research on monkeys learning behavioral chains found that without any timeout period between trials, the animals made a great many errors, largely due to superstitious responses becoming embedded within the reinforced chain. Introducing even brief timeouts between trials substantially reduced these errors.

9PubMed Central. The repeated acquisition of behavioral chains

This finding has practical implications. If you are teaching a multi-step task and the learner keeps inserting odd extra behaviors between steps, like tapping the table before picking up a utensil, adding a brief pause between steps can help clean up the sequence. The pause gives the accidental behaviors a chance to die out instead of being swept along in the reinforcement stream.

Another common mistake is attempting to chain behaviors that the learner cannot yet perform individually. If any link in the chain is weak or absent, the entire sequence breaks down. This is where the shaping-first, chaining-second principle becomes critical. You need solid individual links before you can connect them.

How the Brain Handles Action Sequences

Neuroscience has begun to reveal why chained sequences feel different from individual behaviors once they are well-learned. Research on the basal ganglia, a set of deep brain structures involved in movement and habit formation, suggests that these circuits are important for chunking isolated movements into precise and robust action sequences. During sequence learning, many neurons in the basal ganglia develop activity related to the initiation, execution, and termination of sequences, suggesting that action sequences get processed as unified action units rather than as a series of separate decisions.

10PubMed Central. Shaping action sequences in basal ganglia circuits

This chunking explains a familiar experience: a well-practiced sequence, like tying your shoes or typing your password, flows automatically once you start. But if you are interrupted in the middle, you often have to go back to the beginning. The sequence has been encoded as a chunk, and accessing the middle of it directly is harder than running it from the top. The same research found that plasticity at the connections between the cortex and the striatum (part of the basal ganglia) is critical for this crystallization of action sequences into chunks, and that different pathways within the basal ganglia contribute differently to the process.

11PubMed Central. Shaping action sequences in basal ganglia circuits

This has a training implication. When teaching a chained sequence, the initial learning phase is effortful and step-by-step. But with enough practice, the chain does not stay a chain in the learner’s brain. It becomes a single unit. That transition is partly why overlearning and repeated practice matter even after someone can perform a sequence correctly: the neural representation is still consolidating.

Procedural Skill Training Beyond the Clinic

Shaping and chaining principles show up in skill training far beyond behavioral therapy, even when trainers do not use the terminology. Surgical training, for instance, involves both. A review of training design for complex procedural motor skills, with a focus on laparoscopic surgery, emphasizes how training design choices affect the acquisition, durability, transfer, and efficiency of learned skills. The same factors that matter in a therapy room, like how you break down the task and how you sequence practice, turn out to matter when someone is learning to operate through a tiny incision.

12PubMed. Optimal training design for procedural motor skills: a review and application to laparoscopic surgery

Music instruction works the same way. Learning a difficult piano passage might involve shaping the finger movements for a tricky interval (practicing just that hand position until it is comfortable), then chaining the passage together measure by measure, sometimes backward from the ending so that the player always finishes in well-practiced territory. Sports coaching follows similar logic: a gymnastics routine is a chain of individual skills, each of which may need to be shaped before it can be linked in.

In all these domains, the practical question is the same one faced by a therapist working with a child: does the learner need to build a new response, or connect existing ones? The answer determines which tool to reach for first.

Reward Shaping in Artificial Intelligence

The language of shaping has migrated into machine learning, specifically into reinforcement learning, where “reward shaping” refers to giving an AI agent intermediate rewards that guide it toward a distant goal. Without reward shaping, an agent trying to solve a long, multi-step task faces a sparse reward problem: it only finds out whether it succeeded at the very end, after a potentially enormous number of steps. This is roughly analogous to trying to shape a behavior without reinforcing any approximations along the way.

A recently proposed framework called HAFER addresses this by integrating hierarchical adaptive reward shaping, which dynamically adjusts local skill rewards so they align with the global task objective. The framework improves the efficiency and stability of task execution and enables smooth transitions between skills in complex, long-horizon tasks.

13Neurocomputing. Seamless skill transitions with hierarchical reward shaping and failure-driven replay

The parallel to behavioral chaining is striking. In both cases, the problem is getting a learner, biological or digital, through a long sequence when the only natural reinforcement comes at the end. The solution in both cases is to make intermediate points in the sequence reinforcing in their own right, whether through conditioned reinforcement in a behavioral chain or through engineered reward signals in a computational one. The conceptual debt that modern AI training owes to mid-twentieth-century behavioral research is rarely acknowledged, but it is hard to miss once you look for it.