Small area estimates (SAEs) can provide information about health behaviour at small area levels that is otherwise not available. Because of its increasing use by policy makers, more attention needs to be paid to the reliability of these estimates. This paper reports on smoking prevalence data generated for London at the neighbourhood level using spatial microsimulation modelling. We test the reliability of smoking prevalence estimates at the neighbourhood level using different input datasets. The paper further underlines the importance of estimating health behaviours at the small area level, particularly in diverse cities such as London, where estimation at the city level can mask significant spatial differences.