Spatial granularities allow one to qualify classical data adding them space locations. In order to compare data qualified with different granularities and to associate data to different granularities (e.g., in analysis similar to drill-down and roll-up operations), it is necessary to know how the involved granularities are related. However, the explicit evaluation of these relationships may be heavy from a computational point of view. Moreover, the explicit evaluation of these relationships could not be requested, as relationships can be derived from already established ones. Thus, in this paper, we propose an inference system for deriving spatial relationships that definitely hold, starting from a given set of relationships between spatial granularities, without evaluating them explicitly.