A clustered randomized design was used in which 28 inclusive preschool classrooms were randomly assigned to receive 2 years of training and coaching to fidelity in the LEAP (Learning Experiences and Alternative Program for Preschoolers and Their Parents) preschool model, and 28 inclusive classes were assigned to receive intervention manuals only. In total, 177 intervention classroom children and 117 comparison classroom children participated. Children were similar on all measures at start. After 2 years, experimental class children were found to have made significantly greater improvement than their comparison cohorts on measures of cognitive, language, social, and problem behavior, and autism symptoms. Behavior at entry did not predict outcome nor did family socioeconomic status. The fidelity with which teachers implemented LEAP strategies did predict outcomes. Finally, social validity measurement showed that procedures and outcomes were favorably viewed by intervention class teachers. (Contains 1 figure and 6 tables.)