Mesh simplification and discrete levels of detail (LOD) are wellstudied areas of research in computer graphics. However, until recently, most of the developed algorithms have focused on simplification and viewing of a single object with a large number of polygons. When these algorithms are applied to a large collection of simple models, many objects may be completely erased, leading to results that are misleading to the viewer. In this paper, we present an approach to simplifying city-sized collections of 2.5D buildings based on the principles of "urban legibility" as defined by architects and city planners. We demonstrate that our method, although similar to traditional simplification methods when compared quantitatively, better preserves the legibility and understandability of a complex urban space at all levels of simplification.