Recent experiments indicate the need for revision of a model of spatial memory consisting of viewpoint-specific representations, egocentric spatial updating and a geometric module for reorientation. Instead, it appears that both egocentric and allocentric representations exist in parallel, and combine to support behavior according to the task. Current research indicates complementary roles for these representations, with increasing dependence on allocentric representations with the amount of movement between presentation and retrieval, the number of objects remembered, and the size, familiarity and intrinsic structure of the environment. Identifying the neuronal mechanisms and functional roles of each type of representation, and of their interactions, promises to provide a framework for investigation of the organization of human memory more generally.