Evaluated the effectiveness of a lights-out procedure in reducing off-task behavior in 4 retarded children, aged 3-10 yrs (mean IQ 32). Ss were trained on a lever pressing task as an analog to discrete-trial training procedures. Positive reinforcement sessions were alternated with concurrent escape/avoidance procedures during which lights-out was contingent on nonresponding in an alternating treatment design. The lights-out contingency rapidly and effectively reduced the length of time spent off-task and slightly increased response rate. Educational and therapeutic implications for retarded and autistic children are discussed. (22 ref)