Spatial analysis of criminal incidents is an old and important technique used by crime analysts. However, most of this analysis considers the aggregate behavior of criminals rather than individual spatial behavior. Recent advances in the modeling of spatial choice and data mining now enable us to better understand and predict individual criminal behavior in the context of their environment. In this paper, we provide a methodology to analyze and predict the spatial behavior of criminals by combining data mining techniques and the theory of discrete choice. The models based on this approach are shown to improve the prediction of future crime locations when compared to traditional hot spot analysis.