A randomized experiment of Comer's School Development Program was conducted in 23 middle schools in Prince George's County, Maryland. The school population is predominantly African American, with considerable internal variation in household socioeconomic standing. This study involved repeated measurement with more than 12,000 students and 2,000 staff a survey of more than 1,000 parents, and extensive access to student records. It showed that Comer schools implemented some of the program's central elements better than control schools, but not all or even most of them. This shortfall in program implementation may have been responsible for students in the experimental schools not changing any more than controls. Quasi-experimental analyses showed that the program theory may be correct in many of its predictions about student changes in psychological and social outcomes, but not achievement. However, achievement gains were found in schools with a more explicit academic focus, suggesting that improving this focus should be as central to Comer's program theory as improving a school's social climate. Even more needed, though, are ways to improve program implementability, the sine qua non for student change.