Learning Together is a whole-school intervention, evaluated using a randomized controlled trial in southeast England, which reduced bullying and improved physical and mental health. This paper examines trial data using fuzzy-set qualitative comparative analysis to test hypotheses derived from embedded qualitative research about potential causal pathways. Analyses suggested that the intervention worked via three mechanisms: improving student commitment to school; improving student pro-social skills; and de-escalating conflict and bullying. Evidence also suggests that these mechanisms may have been activated via other resources in schools not receiving Learning Together resources. The analysis suggests which contextual features may be important for activating these mechanisms.