This study investigated the effect of delayed reinforcement on the across-setting generalization of behaviour change. Eight children aged between 11 and 13, members of a special class at an intermediate school, served as subjects. Off-task behaviour was monitored during two classroom lessons: the contingent lesson, performance in which determined subsequent reinforcement, and the generalization lesson, in which no reinforcement contingencies were provided. Two forms of delayed reinforcement: early — delivered immediately following the setting in which the critical behaviour occurred — and late — delivered only after several other settings had been encountered — were sequentially presented in an ABCB design. Off-task behaviour decreased under both reinforcement conditions. However generalization was only evident when the late delayed reinforcement was operating. Results suggest that a temporal delay in the delivery of reinforcement is more likely to lead to generalization of behaviour change than is the delivery of reinforcement immediately following the contingent lesson.