Which data model is best for the spatial information stored in a GIS? This question leads to several recurring debates that consider various parameterizations of the spaces of possible data models and data structures. We list three examples: the long-running raster/vector debate is limited by GIS implementation technology, the tiled/seamless debate is concerned with the practical and conceptual views of a large data set, and the object/layer/field debate appeals to human cognition of space.