Objective: The complexity and scale of health promotion interventions present challenges for the standardization of delivery. Furthermore, health promotion practice favours adapting interventions according to perceived client need. This paper examines the fidelity of intervention delivery within A Stop Smoking in Schools Trial (ASSIST), identifying if and why variations occurred, the consequences of these variations for the integrity of the intervention, and the broader implications for the delivery of evidence-based health promotion in schools. Design: The ASSIST intervention involved influential Year 8 students being trained to encourage their peers, through informal conversations, to be smoke-free. Its effectiveness was evaluated by a pragmatic randomized controlled trial with an integral process evaluation to examine the context, implementation and receipt of the intervention. Setting: Thirty secondary schools in South Wales and the west of England. Method: A variety of methods, including interviews, questionnaires and observation, were used to obtain data from key participants. Qualitative data were analysed using the constant comparative method. Results: Whilst overall fidelity was high, some variations were observed. Despite this variation, the intervention was largely delivered successfully and appropriately across a range of schools. Lessons learned from ASSIST enabled us to develop a model to categorize variations in intervention delivery. Conclusion: The model developed can inform the design of health promotion interventions and identify the level of fidelity which should be expected outside of a trial. We believe that this model can contribute to good practice in the implementation of evidence-based health promotion in schools. (Contains 8 tables and 1 figure.)