In the present study, joint-control training was applied when teaching manded selection responses to children with autism. Four vocal children with autism participated in the first experiment, two males (ages seven and eight) and two females (ages seven and nine). The results showed that it was only after object-word naming was trained under joint control that the symmetrical performance of manded selection responses appeared with no additional training. Four non-vocal children with autism participated in the second experiment, two males (ages six and seven), and two females (ages twelve and thirteen). These results also showed that it was only after the joint tact/self-mimetic/sequelic control training that the symmetrical performance of manded selection responses appeared with no additional training.