Multiple exemplar training (MET) is a teaching strategy used in applied behavior analysis where a skill is practiced across many different examples, settings, people, or materials so the learner can apply that skill flexibly in new situations. Rather than teaching one version of a behavior and hoping it transfers, MET deliberately builds in variety during instruction to promote generalization from the start.
How Multiple Exemplar Training Works
The core idea is straightforward: if you only teach a child to greet one person in one room, there’s no guarantee the child will greet different people in different rooms. MET solves this by rotating through a range of examples during teaching itself. The examples are chosen to reflect the diversity of situations the learner will actually encounter in daily life. When the learner’s behavior comes under the control of the relevant features shared across those examples, encountering a brand-new situation with similar features will naturally trigger the same skill.
Say you’re teaching a child to label “dog.” Instead of showing the same flashcard repeatedly, you’d use photos of different breeds, toy dogs of various sizes, real dogs at the park, and cartoon dogs in a book. The child learns to respond to what makes something a dog (four legs, fur, tail, barks) rather than memorizing one specific image. That’s the mechanism at work: by varying the irrelevant features while keeping the essential features constant, the learner zeros in on what actually matters.
This same logic applies to responses, not just stimuli. If you’re teaching a child to share, you wouldn’t practice sharing only one toy with one peer. You’d vary the toys, the peers, the verbal phrases used to offer sharing, and the settings. Research on children with autism found that training multiple examples of both the objects shared and the verbal responses used during sharing produced generalization to entirely new sharing situations that were never directly taught.
Why Simple Repetition Falls Short
Traditional teaching often follows a pattern: teach one example until the learner masters it, then move to the next. This can produce accurate but rigid performance. A child might perfectly label a red apple on a flashcard but freeze when asked about a green apple on the table. The behavior became tied to specific, narrow stimulus features rather than the broader concept.
The foundational framework for MET comes from a 1977 paper by Stokes and Baer, which reviewed how the field handled generalization and found that most programs simply used a “train and hope” approach. They identified nine strategies for actively programming generalization, and “train sufficient exemplars” was among the most practical. The idea is that after enough varied examples, a tipping point occurs where the learner begins responding correctly to untrained examples without any additional instruction. That tipping point is generalization.
Choosing the Right Examples
The quality of the examples matters as much as the quantity. Effective exemplar sets sample the full range of stimulus features the learner is likely to encounter. If you’re teaching the concept “big,” you’d want to include objects that differ in color, shape, and material but share the critical feature of being relatively large compared to a reference point. You’d also include borderline cases and clear contrasts, so the learner sharpens their discrimination.
There’s no universal magic number for how many exemplars are needed. Some learners begin generalizing after three or four well-chosen examples; others need more. The practical test is whether the learner starts responding correctly to new, untrained examples. Once that happens, you’ve trained “sufficient” exemplars for that particular skill and that particular learner. Practitioners typically probe with novel stimuli after each new exemplar is taught, watching for the moment generalization emerges.
Variation should also happen across dimensions beyond just materials. Changing the instructor, the room, the time of day, and the way instructions are phrased all help ensure the skill isn’t anchored to one narrow context.
MET in Language and Relational Learning
Multiple exemplar training plays a particularly important role in teaching language-related skills. In Relational Frame Theory, a framework for understanding how humans learn to relate concepts to one another, MET is the proposed mechanism by which children develop flexible, abstract thinking. When a child learns that “bigger than” applies across hundreds of comparisons (buildings, animals, numbers, sounds), the relational concept itself becomes generalized. The child can then apply “bigger than” to a comparison they’ve never encountered before.
This extends to skills like reading comprehension, analogical reasoning, and perspective-taking. In each case, the full range of performances results from training a set of exemplars that samples enough variation in both the stimuli presented and the responses expected. Behavior analysts often point to a history of multiple exemplar training to explain how people develop generalized abilities like understanding sarcasm, following novel instructions, or categorizing unfamiliar objects.
Practical Applications
MET is used widely in therapy for children with autism, but the principle applies to any teaching context where you want skills to transfer. Common applications include:
- Social skills: Practicing greetings, turn-taking, or conversation starters with different peers, in different locations, using different phrases
- Daily living skills: Teaching handwashing at home, at school, and in public restrooms so the routine isn’t dependent on one specific sink
- Academic concepts: Introducing math operations with varied number sets, word problems, and physical manipulatives rather than drilling one worksheet format
- Communication: Teaching requests using different items, different listeners, and different contexts so the learner initiates communication spontaneously rather than only on cue
The implementation follows a general pattern. First, identify the target skill and the range of conditions under which it should occur. Then select a diverse set of teaching examples that represent that range. Teach using those examples, rotating them within and across sessions rather than drilling one at a time to mastery. Periodically test with brand-new examples to see if generalization has occurred. If the learner responds correctly to untrained examples, the exemplar set was sufficient. If not, add more varied examples and continue.
How MET Differs From Other Approaches
MET is sometimes confused with simple mass practice or drilling, but the distinction is important. Drilling involves repeating the same example until it’s automatic. MET involves rotating through different examples to build flexibility. The goal of drilling is fluency with a specific response; the goal of MET is generalization to untrained situations.
It also differs from “training loosely,” another generalization strategy identified by Stokes and Baer. Training loosely means being deliberately less precise about how instruction is delivered, allowing natural variation to creep in. MET is more systematic: the variation is planned and the exemplars are deliberately selected to represent the range of conditions the learner will face. Both strategies promote generalization, but MET gives the practitioner more control over which features are varied and how.
The strength of MET is its reliability. Rather than hoping a skill transfers, you’re engineering the conditions for transfer directly into the teaching process. The limitation is that it requires thoughtful planning. Poorly chosen exemplars, ones that are too similar to each other or that don’t represent real-world conditions, can waste instructional time without producing generalization. The diversity of the example set, not just its size, is what drives the outcome.