Most crowd simulators focus on navigation and agents flow. In this paper, we present another perspective that concentrates on the overall distribution of virtual agents and uses psychological preferences for choosing goal locations. Both observation and published theory indicate that most people prefer to maintain their personal space as event spaces increase in density, particularly when they have no previous relationship to other individuals. The geometric structure that naturally forms could be highly approximated by a Voronoi tessellation. Our method allows users to specify sub-regions of an environment and tag the regions with information (e.g., permitted densities and features). A Centroidal Voronoi Tessellation (CVT) is then automatically constructed over the entire virtual world. The centers of mass of each resulting cell are taken as potential agent goal locations. Individual virtual agents then have the ability to choose their preferred goal locations on the basis of their own characteristics (e.g., personality traits, needs, and interests), the CVT, and semantic features of the sub-regions. This method results in more meaningful crowd simulations with minimal additional user effort. Copyright © 2012 John Wiley & Sons, Ltd.