In three experiments, we compared the effects of instructional arrangements that varied in: teacher versus peer mediators, methods used, levels of student academic responding generated, and content taught and tested. Instructional arrangements (i.e., tasks, structure, teacher position, teacher behavior) and students' levels of academic responding were measured by an observation system which served as an index of the independent variables. Students' accuracy on weekly spelling, arithmetic, and vocabulary tests and pre- and post-standardized achievement tests (Experiments 2 and 3 only) were the dependent variables. Results indicated that the classwide peer tutoring, compared to the teacher's procedure, produced more student academic responding and higher weekly test scores, regardless of treatment order or subject matter content (Experiment 1). The four lowest performing students in each class, in particular, benefited from peer tutoring, often performing as well as the other students. These findings were replicated in Experiments 2 and 3 wherein content taught/tested was also manipulated. Standardized test score gains were higher in those areas in which peer tutoring was used longest. Issues related to the functional analysis of instruction and achievement gain are discussed.