In two independent parent training projects (Kansas and Mississippi), mothers of deviant young children were observed to follow almost all child behaviors with attention. The mothers were then trained to use differential attention procedures to increase their child's appropriate behaviors and to decrease deviant behaviors. Contrary to expectations, the differential attention procedure produced substantial increases in deviant behavior for four of the children. This adverse effect was maintained over many sessions and was replicated in single organism, reversal designs. A fifth child showed no change. A sixth child showed some improvement. However, this effect was not recovered in a second application of differential attention, and the child became worse. The results underline the importance of subject generality in applied behavior analysis and strongly suggest that service programs using operant techniques must carefully evaluate their effects on behavior.