This article describes an application of genetic algorithms to the analysis of spatially referenced data. A genetic algorithm is used to refine the specification of an hedonic regression model of spatially distributed residential property prices. The process of refinement concerns the search for good definitions of spatially defined variables. The fitness function for the genetic algorithm is provided by the coefficient of determination of the model. The regression results produced by the refined model are compared with those produced by a model containing a set of spatially defined variables based on information provided by an expert on local property prices.