In interventions attempting to remediate deficiencies in the skills repertoire of developmentally delayed children, no less than in medical interventions, it may be fairly said that less is more. That is, the instructor should intervene as little as possible both from the perspective of efficient instructional practice and from time allotment concerns which modern classrooms face. Evidence from this laboratory has indicated that in skills training for children with severe developmental delays the passive observation of a model demonstrating the target skill is more effective than interactive modeling involving hand-over-hand instruction with verbal prompting. We have considered the role of verbal prompting in interactive modeling and have found that prompts intended to provide typical social reinforcers are counterproductive (e.g. Biederman, Davey, Ryder, & Franchi, 1994). The present study examines the efficacy of hand-over-hand modeling with response-contingent verbal prompts. In such instruction, tasks are divided into identifiable sequential components, and the achievement of each component is marked by the delivery of some form of verbal prompt. In a within-subjects design, children were trained in one skill with response-contingent verbal prompts and in a second skill with simple passive observation. A separate group of children were trained with less rigorous verbal prompting in one skill and with passive observation in a second. Consistent with previous research, we found that passive modeling was overall significantly more effective than hand-over-hand modeling and moreover that passive modeling was significantly more effective than hand-over-hand modeling with response-contingent prompting. Our evidence therefore indicates that current classroom practice in training basic skills to children with severe developmental delays may require reassessment in that simple observation of modeled skills appears to be more effective than more labor-intensive instruction.