We analyze the impact on high-stakes assessments of incentivizing students’ effort in a field experiment with over 10,000 high school students. We contribute to the literature by using our rich data and machine learning techniques to explore treatment heterogeneity; by comparing financial and non-financial rewards in rewarding effort rather than grades; and by using high-stakes outcomes. We find little average impact of incentives in the overall population, but we identify a “right tail” of highly responsive students: in the upper half of the responsiveness distribution, test scores improve by 0.1-0.2 SD, about half the attainment gap between poor and non-poor students.