Research on teaching STEM, especially in the areas of teaching coding for students with ASD, is lacking. The purpose of the current study was to evaluate the effects of using model-lead-test to teach an elementary-aged student with ASD and severe problem behavior the following dependent variables: (a) calibrating the robot; (b) drawing tracks for the robot to follow; and (c) creating a code (e.g., to make the robot move quickly). Results of the multiple probe across skills design demonstrate a functional relation between the model-lead-test strategy and the acquisition of all of the skills. Further, he generalized the coding skill to novel codes, and maintained the skills over time. Implications, study limitations, and recommendations for future research are discussed.