The authors' purpose in this study was to compare the effectiveness of two instructional approaches on mildly handicapped and nonhandicapped students' science achievement. Students were assigned at random to one of two conditions: (a) direct instruction, and (b) discovery teaching. The content of the lessons remained constant across conditions and focused on such concepts as displacement, flotation. variable, controlled experimentation, and scientific prediction. The results show that students in both groups learned equally well as measured by an immediate posttest. However, students in the discovery teaching condition outperformed their direct instruction counterparts on a retention test administered two weeks after the posttest. Finally, learning-disabled students in the discovery condition performed better than their direct-instruction counterparts on a performance-based measure designed to assess generalization. Implications for research and for practice are discussed.