BackgroundConduct problems, which include aggression, lying, and stealing, can differ in terms of continuity and desistence across different stages of life, such as childhood and adolescence. Although several research efforts have investigated individual-level factors associated with continuity and desistence of conduct problems in adolescence, less is known about school-level factors that might underlie such patterns. MethodsWe undertook longitudinal latent class analysis to identify trajectories of conduct problems across adolescence (age 11–15 years) using the Learning Together study dataset (n=3901, 50·7% female), which includes adolescents in 40 state schools within Greater London, UK. We then investigated the role of several school-level factors (such as Ofsted rating, gender of students, school deprivation, school type, and school atmosphere, according to information gathered through the government website or through self-reported questionnaire) in predicting class membership using multinomial logistic regression. On the basis of the Bayesian Information Criterion and entropy, we selected the best fit model with two classes of conduct problems using three data points: a stable low and a moderate-high class across male and female students. A multi-level model was employed to account for school-level clustering. FindingsSeveral school-level factors predicted persistent patterns of conduct problems in male and female students separately, even after controlling for family-level confounders such as socio-economic status and family structure. However, positive school atmosphere was the factor that was most strongly associated with reduced risk of persistent conduct problems across male and female students. More specifically, a better school atmosphere at baseline was associated with lower risk of being in the moderate-high class for conduct problems in both female (OR 0·365, 95% CI 0·253–0·527) and male students (0·290, 0·186–0·451). InterpretationRelationships between teachers and students, sense of belonging to the school, and participation in school activities are aspects that researchers and schools should consider when implementing prevention and intervention programs for youth with conduct problems and antisocial behaviour. A strength of this study was that it included a large sample of adolescents. A limitation was that we did not investigate causality, which should be addressed in future research. FundingPartly supported by the NIHR through a funded intervention study called Learning Together (ISRCTN10751359).