The authors investigate geographical patterns of homicide in Sao Paulo, Brazil. The geography of crime in developing world cities has been an underresearched area in part because of the lack of good-quality, geocoded offence data. In the case of Sao Paulo the availability of a new digital police dataset has provided the opportunity to improve our understanding of its crime patterns. The authors report the testing of hypotheses about the spatial variation in homicide rates. This variation is explained by poverty, situational conditions determined by differences in land use, and processes that indicate links with the geography of drug markets and the availability of firearms.