We summarize two lines of research that investigated whether human spatial memories are viewpoint dependent (e.g., viewer-centered reference frames) or viewpoint independent (e.g., scene-centered reference frames). In one series of experiments, participants made judgments of relative direction after viewing a spatial layout from one or two perspectives. The findings indicated that multiple views of a spatial layout produced multiple viewpoint dependent representations in memory. These findings were corroborated by the results of experiments on scene recognition. These experiments showed, again, that multiple views of a scene produced multiple viewpoint dependent representations in memory, and that a novel view of a familiar scene was recognized by normalizing it to the most similar view in memory. A preliminary computational model of scene recognition formalizes several of these concepts.