This paper analyzes data from Project STAR, a randomized experiment in Tennessee that assigned 11,600 kindergarten students and teachers to small classes (13-17 students), regular-size classes (22-25 students), or regular-size classes with a teacher's aide. The study found that: 1. On average, students in small classes performed about 4 percentile points higher in their first year, regardless of the grade in which they first attended a small class. 2. After the initial assignment to a small class, students' performance increased by about 1 percentile point per year relative to those in regular-size classes. 3. Teacher aides had little effect on student achievement. 4. Class size had a larger impact on test scores for minority students and those on free lunch. 5. The beneficial effect of smaller classes did not appear to be due to Hawthorne effects. The paper also discusses the limitations of the experiment, such as re-randomization in regular-size classes and nonrandom transitions between classes, and addresses the robustness of the findings to these limitations. The analysis uses various statistical methods, including OLS and Two-Stage Least Squares (2SLS), to estimate the effects of class size on student achievement. The results suggest that the initial effect of attending a small class is highly significant, while the cumulative effect is marginal. The paper also examines the impact of attrition and heterogeneous treatment effects, finding that nonrandom attrition did not bias the estimated class size effects.This paper analyzes data from Project STAR, a randomized experiment in Tennessee that assigned 11,600 kindergarten students and teachers to small classes (13-17 students), regular-size classes (22-25 students), or regular-size classes with a teacher's aide. The study found that: 1. On average, students in small classes performed about 4 percentile points higher in their first year, regardless of the grade in which they first attended a small class. 2. After the initial assignment to a small class, students' performance increased by about 1 percentile point per year relative to those in regular-size classes. 3. Teacher aides had little effect on student achievement. 4. Class size had a larger impact on test scores for minority students and those on free lunch. 5. The beneficial effect of smaller classes did not appear to be due to Hawthorne effects. The paper also discusses the limitations of the experiment, such as re-randomization in regular-size classes and nonrandom transitions between classes, and addresses the robustness of the findings to these limitations. The analysis uses various statistical methods, including OLS and Two-Stage Least Squares (2SLS), to estimate the effects of class size on student achievement. The results suggest that the initial effect of attending a small class is highly significant, while the cumulative effect is marginal. The paper also examines the impact of attrition and heterogeneous treatment effects, finding that nonrandom attrition did not bias the estimated class size effects.