We estimate the causal impact of school and classroom gender composition on achievement. We take advantage of the random assignment of Korean middle school students to single-sex schools, coeducational (coed) schools with single-sex classes, and coed schools with mixed-gender classes. Male students attending single-sex classes within coed schools score 0.10 of a standard deviation below male students in mixed-gender classes, and this achievement gap is entirely accounted for by classroom gender composition. Conversely, male students attending single-sex schools outperform their counterparts in mixed-gender classes by 0.15 of a standard deviation. The significant impact of single-sex schools on male students’ achievement are not driven by classroom gender composition, but largely accounted for by increases in student effort and study-time. We find little evidence that classroom or school gender composition affect the outcomes of female students.