Background: Embedded instruction and discrete-trial training are both recommended for teaching children with autism, but there is little research available comparing the two. The present study compared embedded instruction with discrete-trial training for a 12-year-old boy with autism. Method: An initial functional analysis indicated that the boy's self-injury was maintained by escape from task demands. Instructional sessions to teach adaptive behaviours were conducted under two conditions: (i) during embedded instruction, learning trials were inserted into ongoing activities at a rate of approximately 1.5 per minute; (ii) during discrete-trial training, instructional opportunities were incorporated into structured sessions at a rate of 4 per minute. In both conditions, the system of least prompts was used to teach relevant target responses. Effects of the two teaching formats were evaluated using an ABABA design. Results: Higher rates of self-injury and fewer correct responses occurred during discrete-trial training. Mood ratings were also lower during discrete-trial training. Conclusions: The results suggest that self-injury, correct responding, and mood may be sensitive to the type of instructional format. Although discrete-trial training can be highly effective, it may be preferable to start with embedded instruction when the child presents with self-injurious escape behaviour. (Contains 3 figures.)