Despite rising demand, evidence regarding the effectiveness of social-emotional learning (SEL) programs implemented at scale in low- and middle-income countries remains limited. Building on an earlier randomized control trial showing largely null program effects in year one, the present study examines the effectiveness of a school-wide SEL intervention during its second year of implementation in Rio de Janeiro, Brazil. The study sample included 3,188 third- and fourth-grade students attending 84 public primary schools during the 2018 school year. We identified small (d = 0.09 − 0.12) iatrogenic impacts of school-level random assignment to the intervention on students’ behavior problems, ability to name appropriate reactions to emotional situations, and working memory during year two, and no average impacts on students’ labeling of emotional facial expressions or inhibitory control. Two-stage least squares analyses revealed that students whose teachers implemented more program lessons during year two experienced more negative outcomes. Using exploratory analyses of impact heterogeneity and qualitative feedback from study teachers, we discuss possible reasons for these unexpected findings and provide recommendations for supporting children’s social-emotional needs in diverse settings.