Summary: Within this article we are dealing with the issue of spatial problem solving, this means referring to a collection of data placed in a geometrical space. We first of all describe the general framework, before presenting our novel approach to this type of problems based on the use of Reactive Multi-Agent Systems. We are developing this paradigm, we give a formal mathematical definition to it and we illustrate on an example in image analysis. Starting with empirical observations, we are then presenting a methodology to build such systems based on the notion of interaction constraints, and detail how the initial specifications of the topological and morphological constraints may guide the system towards a solution. As a conclusion, we are instantiating our work on two types of spatial problems: multi-criteria spatial decision support systems and cartographic generalisation.