In order to describe a spatial environment, people must take a perspective on it. Previous researchers had claimed that in describing space, speakers take listeners on mental tours, using a consistent perspective. In contrast, we find that people use survey and mixed perspectives as well as route perspectives, and that the configuration of an environment affects perspective choice. We show that gaze, route, and survey perspectives use language differently and argue that they correspond to prototypic relative, intrinsic, and extrinsic frames of reference, respectively. We speculate that the correlation of viewpoint, referent, and terms of referent in the three perspectives occurs because each reflects a natural way of interacting with an environment.