How do people learn the spatial relationship between places that share no common landmarks (e.g., a windowless interior room & the outside world)? A potential solution is navigation between these environments. For example, if one can keep track of the orientation of one environment while walking toward another environment, then she can establish the spatial relationship between these environments when she arrives at the second environment. Two experiments investigated how people learn the directional relationship between environments through navigation. Participants walked from a windowless interior room to a testing position outside the building, along a path with six 90 deg turns, and then returned to the room along the same path. People failed to learn the directional relationship between the room and the outside world, because they lost track of the room orientation when they walked outside, and lost track of the orientation of the outside world when they returned to the room. When participants were forced to report the orientation of the room after each turn along the path (i.e., forced to “update” the room orientation), they successfully kept track of the room orientation after they walked outside, and thus reported the geographical orientation of the room in correct relationship to the outside world. However, when they returned to the room, again reporting the room orientation along the way, they failed to report the orientation of the outside world in correct relationship to the room. These results suggest that spatial learning through navigation requires updating of the proper target (i.e., the environment one just left), which does not occur automatically in everyday navigation, and updating of the wrong target can override spatial relationships already learned.