This paper investigates the effect of class-size reduction on students’ academic outcomes, with a particular emphasis on its heterogeneity based on classmates’ characteristics. We estimate the causal effects of class-size reduction on students’ mathematics and language test scores by controlling for student and teacher fixed effects. To address potential endogeneity, we employ the predicted class size with a cap as an instrumental variable for the actual class size. Utilizing rich panel data on Japanese primary school students, our findings indicate a positive and robust average effect of class-size reduction on mathematics test scores. Furthermore, we find that classes with high-ability classmates benefit even more from class-size reduction in terms of language test scores. The effect of class-size reduction on mathematics test scores is found to depend positively on the ability of the lowest-achieving student in a class. Additionally, classes with a higher proportion of female students tend to benefit more from class-size reduction. Our results lend support to the theoretical framework proposed by Lazear (2001).