We tested 128 adults; male and female ages 19 to 56, for the emission of mirrored or non-mirrored responses in generalized imitation (GI). We found that the adults emitted both. Thus, the emission of non-mirrored responses was not part of the criterion for GI in Experiment 2, where we isolated the effects of using a mirror for training imitative responses on the emergence of GI by children with autism (six children, 3-4 years old). We used an experimental-control group design with a “nested” non-concurrent multiple probe design across participants in the experimental group. Test scores and developmental measures were used to match participants who were then assigned randomly to groups. The numbers of instructional trials were yoked between pairs controlling the amount of instruction. GI emerged for the individuals who received the mirror training, but not for those trained face-to-face. One possibility is that mirror training rotating different responses may result in a class of responding whereas, face-to-face training results in specific responses, which may explain why prior studies were unable to induce GI with typically developing young children. Moreover, given the findings herein, the correspondence between seeing and doing as conditioned reinforcement for GI remains a viable theory.