BACKGROUND: We sought to determine whether the Good School Toolkit-Primary violence prevention intervention was associated with reduced victimisation and perpetration of peer and intimate partner violence four years later, and if any associations were moderated by sex and early adolescent: family connectedness, socio-economic status, and experience of violence outside of school. METHODS: Drawing on schools involved in a randomised controlled trial of the intervention, we used a quasi-experimental design to compare violence outcomes between those who received the intervention during our trial (n = 1388), and those who did not receive the intervention during or after the trial (n = 522). Data were collected in 2014 (mean age 13.4, SD 1.5 years) from participants in 42 schools in Luwero District, Uganda, and 2018/19 from the same participants both in and out of school (mean age 18, SD: 1.77 years). We compared children who received the Good School Toolkit-Primary, a whole school violence prevention intervention, during a randomised controlled trial, to those who did not receive the intervention during or after the trial. Outcomes were measured using items adapted from the International Society for the Prevention of Child Abuse and Neglect Child Abuse Screening Tool-Child Institutional. We used mixed-effect multivariable logistic regression, with school fitted as a random-effect to account for clustering. RESULTS: 1910 adolescents aged about 16-19 years old were included in our analysis. We found no evidence of an average long-term intervention effect on our primary outcome, peer violence victimization at follow-up (aOR = 0.81, 95%CI = 0.59-1.11); or for any secondary outcome. However, exposure to the intervention was associated with: later reductions in peer violence, for adolescents with high family connectedness (aOR = 0.70, 95% CI 0.49 to 0.99), but not for those with low family connectedness (aOR = 1.07, 95% CI 0.69 to 1.6; p-interaction = 0.06); and reduced later intimate partner violence perpetration among males with high socio-economic status (aOR = 0.32, 95%CI 0.11 to 0.90), but not low socio-economic status (aOR = 1.01 95%CI 0.37 to 2.76, p-interaction = 0.05). CONCLUSIONS: Young adolescents in connected families and with higher socio-economic status may be better equipped to transfer violence prevention skills from primary school to new relationships as they get older. TRIAL REGISTRATION: Clinicaltrials.gov, NCT01678846, registration date 24 August 2012. Protocol for this paper:  https://www.researchprotocols.org/2020/12/e20940 .