<p>Individuals with a diagnosis of autism are often reported to have challenges recognizing emotions in themselves and others, and many teaching programs for emotional recognition utilize facial expressions as an emotional indicator of what is felt. Although facial expression is an important component, research suggests emotions are not universally expressed. Initial steps in teaching emotion recognition may require teaching both the shared meaning of the emotion words used within a given environmental context and the effects these words produce on the verbal community. The present study employed a concept analysis of “fear” and “anger,” from which rational sets of examples and non-examples were created. Four cis-gender, adolescent, male learners diagnosed with autism were shown computer animations of examples and non-examples of fear and anger and asked to provide the emotion word and describe the corresponding contingency. A test-revise-test control analysis was used to make iterative changes in the program until the learners emitted correct responses when shown novel exemplars. Baseline and test scores indicate that learner behavior was controlled by the instructional program, demonstrating learners identified and tracked changes in the depicted environments that led to descriptions of fear and anger. This study illustrates how data-driven refinements and instructional design methodologies can lead to carefully sequenced, individualized, efficient teaching procedures for individuals with autism.</p>