Evidence on optimal school size is mixed. We estimate the effect of transitory changes in school size on the academic achievement of fourth-and fifth-grade students in North Carolina using student-level longitudinal administrative data. Estimates of value-added models that condition on school-specific linear time trends and a variety of teacher-by-school, student, and school-by-year fixed effects suggest that, on average, there is no causal relationship between school size and academic performance. However, two subgroups of interest are significantly harmed by school size: socioeconomically disadvantaged students and students with learning disabilities. The largest effects are observed among students with learning disabilities: A 10-student increase in grade size is found to decrease their math and reading achievement by about 0.015 test-score standard deviations.