Abstract : New information technologies allow more and more access to geographical information for a wide variety of public. Convivial and intuitive interface needing quite no training are available. Nevertheless, these tools have a limited power of expression when very precise specification of the information is needed. The ``geographical anteserver'' gives a complete environment allowing the expression of a query in a bi-modal manner (sketch + textual labels). In a first part, we show the two main qualities of the sketch modality: no need of training and great expressiveness. As a counterpart, we point the ambiguities due to the different levels of interpretation that such a modality leads to. The second part describes the use of the multimodal interface and focus on the complementarity of text and sketch to disambiguate the query thanks to a controlled interaction with the end user using a mechanism of sliding of meaning. Textual interpretation of a sketch and sliding of meaning require a fine spatial analysis. At this point, we present the heart of this research : a qualitative spatial analysis model using recurrently the 9-intersections model of M.J. Egenhofer. This spatial analytical model may capture topological, metric or orientation constraints, contrary to the models presented in the literature. This unification allow an easy implementation of the mechanism of sliding of meaning. As a conclusion, we present our software implementation.