In previous research (D. J. Bryant, B. Tversky, & N. Franklin, 1992; N. Franklin & B. Tversky, 1990), the authors showed that spatial knowledge conveyed by descriptions and direct experience induces participants to take the perspective of a character surrounded by objects. In this study, the authors used models and diagrams to convey the same information. With models, as with descriptions and experience, participants adopted the character's perspective (the spatial framework analysis). With diagrams, participants took an outside perspective (the intrinsic computation analysis). Even when informationally equivalent, different depictions made salient different aspects of the world. When instructed, however, participants were able to take either the inside or the outside perspective in memory for both diagrams and models. Depth cues in depictions also govern participants' perspective. When diagrams contained rich pictorial depth cues, participants used the spatial framework analysis, and when models were viewed without access to depth cues, participants relied on the intrinsic computation analysis. People's knowledge of the world comes not only directly, from experiencing the world, but also indirectly, from descriptions and depictions of the world. Perhaps because of its significance, spatial knowledge has been conveyed by external representations since prehistory. Maps, whether from stone, clay, wood, bark, or paper, have been invented by many cultures (e.g., Brown, 1949; Wilford, 1981). Spatial language alone can act like a map, effectively conveying spatial relations and relative distances (e.g., Bryant, Tversky,