Health interventions increasingly rely on formative qualitative research and social marketing techniques to effect behavioural change. Few studies, however, incorporate qualitative research into the process of program evaluation to understand both impact and reach: namely, to what extent behaviour change interventions work, for whom, in what contexts, and why. We reflect on the success of a community-based hygiene intervention conducted in the slums of Kathmandu, Nepal, evaluating both maternal behaviour and infant health. We recruited all available mother-infant pairs (n = 88), and allocated them to control and intervention groups. Formative qualitative research on hand-washing practices included structured observations of 75 mothers, 3 focus groups, and 26 in-depth interviews. Our intervention was led by Community Motivators, intensively promoting hand-washing-with-soap at key junctures of food and faeces contamination. The 6-month evaluation period included hand-washing and morbidity rates, participant observation, systematic records of fortnightly community meetings, and follow-up interviews with 12 mothers. While quantitative measures demonstrated improvement in hand-washing rates and a 40% reduction in child diarrhoea, the qualitative data highlighted important equity issues in reaching the ultra-poor. We argue that a social marketing approach is inherently limited: focussing on individual agency, rather than structural conditions constraining behaviour, can unwittingly exacerbate health inequity. This contributes to a prevention paradox whereby those with the greatest need of a health intervention are least likely to benefit, finding hand-washing in the slums to be irrelevant or futile. Thus social marketing is best deployed within a range of interventions that address the structural as well as the behavioural and cognitive drivers of behaviour change. We conclude that critiques of social marketing have not paid sufficient attention to issues of health equity, and demonstrate how this can be addressed with qualitative data, embedded in both the formative and evaluative phases of a health intervention.