Five children with autism were taught to match printed words to corresponding pictures. Participants' speed of learning was compared across three training conditions, each involving a 5-s delay of reinforcement, using a within-participants alternating treatments design. In the cue-value condition, a verbal phrase of approval (e.g., "good!") was delivered only after correct responses and again after a 5-s delay when a primary reinforcer was delivered; in the response-marking condition, an attention-eliciting verbal cue (e.g., "look!") was delivered after both correct and incorrect responses, but not prior to the primary reinforcer; in the delay only condition, there were no cues during a 5-s delay. Performance in the no-cue control was inferior to both the cue-value and response-marking conditions, but there was little difference between the latter two conditions. The implications of these results for facilitating learning in applied settings are discussed.