This study was designed to assess the transfer of treatment gains of autistic children across settings. In the first phase, each of 10 autistic children learned a new behavior in a treatment room and transfer to a novel extra-therapy setting was assessed. Four of the 10 children showed no transfer to the novel setting. Therefore, in the second phase, each child who failed to transfer participated in an analysis of stimulus control in order to determine the variables influencing the deficit in transfer. Each of the four children who did not transfer were selectively responding to an incidental stimulus during the original training in the treatment room. Utilizing a reversal design, each of the four children responded correctly in the extra-therapy setting when the stimulus that was functional during training was identified and introduced into the extra-therapy setting. The extreme selective responding and the resulting bizarre stimulus control found are discussed in relation to the issue of setting generality of treatment gains.