One has often to deal with large quantities of data in robotics, either coming from sensors or from background knowledge. Background knowledge, with attached semantics, are usually modeled logically, and sensor data, due to uncertainties concerning their nature, are modeled probabilistically. In this paper we present a scalable method for spatial mapping of indoor environments, through the use of a probabilistic ontology. Reasoning with this ontology allows segmentation and tagging of sensor data acquired by a robot during navigation. We report experiments with a real robot to validate our approach, thus moving closer to the goal of integrating mapping and semantic labeling processes.Sociedad Argentina de Informática e Investigación Operativa