The role of context was examined in the selection and integration of independently learned spatial relationships. Using a dynamic 3D virtual environment, participants learned one spatial relationship between landmarks A and B which was established in one virtual context (e.g., A is left of B) and a different spatial relationship which was established in another virtual context (e.g., A is in front of B). A spatial relationship between B and a goal was then established, using the same landmark-to-goal relationship in each context. When given A in either context at test, participants were more likely to choose a position in expectation of the goal, based on the relationship of A to B previously established in that context. Thus, context can be used to disambiguate conflicting spatial information. This research is discussed in light of Miller and Escobar's (2002) retrieval model. (Contains 3 figures.)