The future mass deployment of pervasive and dense sensor network infrastructures calls for proper mechanisms to enable extracting general-purpose data from them at limited costs and in a compact way. The approach presented in this paper relies on a simple algorithm to let a sensor network self-organize a virtual partitioning in correspondence of spatial regions characterized by similar sensed patterns, and to let distributed aggregation of sensorial data take place on a per-region basis. This makes it possible to perceive the network as if it were composed of a limited number of virtual macro sensors, a feature which promises to be very suitable for a number of incoming usage scenarios.