Behaviour change is notoriously difficult to initiate and sustain, and the reasons why efforts to promote healthy behaviours fail are coming under increasing scrutiny. To be successful, health interventions should build on existing practices, skills and priorities, recognise the constraints on human behaviour, and either feature community mobilisation or target those most receptive to change. Furthermore, interventions should strive to be culturally compelling, not merely culturally appropriate: they must engage local communities and nestle within social and ecological landscapes. In this paper, we propose a social ecology perspective to make explicit the links between intention to change, actual behaviour change, and subsequent health impact, as relating to both theory-based models and practical strategies for triggering behaviour change. A social ecology model focuses attention on the contexts of behaviour when designing, implementing or critically evaluating interventions. As a case study, we reflect on a community-directed intervention in rural Gambia designed to reduce malaria by promoting a relatively simple and low-cost behaviour: repairing holes in mosquito bednets. In phase 1, contextual information on bednet usage, transactions and repairs (the 'social lives' of nets) was documented. In phase 2 (intervention), songs were composed and posters displayed by community members to encourage repairs, creating a sense of ownership and a compelling medium for the transmission of health messages. In phase 3 (evaluation), qualitative and quantitative data showed that household responses were particularly rapid and extensive, with significant increase in bednet repairs (p<0.001), despite considerable constraints on human agency. We highlight a promising approach-using songs-as a vehicle for change, and present a framework to embed the design, implementation and critical evaluation of interventions within the larger context-or social ecology-of behaviour practices that are the bedrock of health interventions.