This article investigates the benefits of girls-only classroom instruction in math and science during Grades 9 and 10, in the context of a public coeducational high school. It is based on a longitudinal investigation with 786 participants: 85 girls in all-girl classes, and 319 girls and 382 boys in a regular coeducational program. Preexisting achievement, background, and psychological characteristics were included as covariates to ensure comparability of the groups. Significant post-intervention program effects were found for math and science achievement and course enrollment. In contrast, there were no significant program effects for perceived math competence or math anxiety. Although those psychological characteristics predicted performance, they were independent of program effects (i.e., they did not mediate the program effects). (Contains 3 figures, 6 tables and 1 note.)