This paper examines neighbourhood effects on health within a large Canadian city--Montréal. Our approach is to consider that individual health outcomes are determined both by individual and neighbourhood characteristics and we consciously take on the problem of neighbourhood definition by developing 'natural' neighbourhoods. Our data come from the Montréal health region sample of the 2000/1 Canadian Community Health Survey, a comprehensive national survey that contains information on health outcomes as well as behavioural and socio-demographic information. Respondents were placed into 'naturally' defined neighbourhoods as opposed to arbitrary geostatistical units, responding to calls in the literature to develop more meaningful units of analysis. We also compare the 'natural' neighbourhood approach with the use of standard census tracts as the unit of analysis. Results show significant between-neighbourhood variation in health status with about 3% of the variance in the Health Utilities Index captured at the neighbourhood level, even after controlling for a variety of socio-demographic and behavioural variables at the individual level. Models using census tracts as the unit of analysis had remarkably similar results to the 'natural' neighbourhood models, suggesting that census tracts are good proxies for natural neighbourhood boundaries in studies of neighbourhood effects on health.