Poor adolescent mental health calls for universal prevention. The Mental Health Foundation's 'Peer Education Project' equips older students ('peer educators') to teach younger students ('peer learners') about mental health. The peer-led lessons cover defining good and bad mental health, risk and protective factors, self-care, help-seeking and looking after one another. While previous pre-post evaluations have suggested effectiveness, the mechanisms through which the intervention improves mental health literacy remain unclear. We purposively recruited seven secondary schools across England from 2020 to 2022 and collected data through five observations, 12 staff interviews and 15 student focus groups (totalling 134 students; 46 peer educators aged 14-18 years and 88 peer learners aged 11-13 years). Our realist analysis adopted retroductive logic, intertwining deductive and inductive approaches to test the initial programme theory against insights arising from the data. We developed Context-Mechanisms-Outcome configurations related to four themes: (i) modelling behaviours and forming supportive relationships, (ii) relevant and appropriate content, (iii) peer educators feeling empowered and (iV) a school culture that prioritises mental health support. Our refined programme theory highlights key mechanisms, contexts conducive to achieving the outcomes and ways to improve training, recruitment and delivery to maximise effectiveness for similar peer-led initiatives.