Spatial data mining fulfils real needs of many geomatic applications. It allows the geomatics community to take advantage of the growing availability of geographically referenced data and benefit from this rich information resource. This article addresses spatial data classification for using decision trees. A new method called SCART which differs from conventional decision trees by considering the specifics of geographical data, namely their organisation in thematic layers, and their spatial relationships is proposed. SCART is an extension of CART methods in two ways. On the one hand, the algorithm considers several thematic layers as in the so-called relational data mining area, and on the other hand, it extends discriminating criteria to address concerns about the neighbourhood. As such, the algorithm determines which combination of attribute values and spatial relationships of neighbouring objects provide the best criterion.