This paper evaluates the impact of three different performance incentive schemes using data from a social experiment that randomized 88 Mexican high schools with over 40,000 students into three treatment groups and a control group. Treatment 1 provides individual incentives for performance on curriculum-based mathematics tests to students only, treatment 2 to teachers only, and treatment 3 gives both individual and group incentives to students, teachers, and school administrators. Program impact estimates reveal the largest average effects for treatment 3, smaller impacts for treatment 1, and no impact for treatment 2.