Does relaxing strict school discipline improve student achievement, or lead to classroom disorder? We study a 2012 reform in New York City public middle schools that eliminated suspensions for non-violent, disorderly behavior. Math scores of students in more-affected schools rose by 0.05 standard deviations over three years relative to other schools. Reading scores rose by 0.03 standard deviations. Only a small portion of these aggregate benefits can be explained by the direct impact of eliminating suspensions on students who would have been suspended under the old policy. Instead, test score gains are associated with improvements in school culture, as measured by the quality of student-teacher relationships and perceptions of safety at school. We find no evidence of trade-offs between students, with students benefiting even if they were unlikely to be suspended themselves.