We present an approach to spatial reasoning that is based on homogenous coordinate systems and their transformations. In contrast to qualitative approaches, spatial relations are not represented by symbolic expressions only but additionally by parameters with constraints, which are subsets of real numbers. Our work is based on the notion of mental models in text understanding introduced in cognitive science. That is, we model the understanding of descriptions of spatial configurations by constructing a representation of the class of situations that are compatible with the description. Within our approach a spatial relation between two objects is represented by constraining the position of one of the objects with respect to the reference frame of the other one. That is, inferences of spatial relations not given explicitly in a text depend on the frame of reference which is presumed by the (human or computer) system. In this paper we describe a general framework for spatial reasoning based on the notion of mental models and we present some empirical results concerning the influence of the selected reference frame on the inference process.