A new technique for using receptor surface models in quantitative structure-activity relationship (QSAR) analysis is described. Receptor surface models provide compact, quantitative descriptors which capture three-dimensional information about putative receptor/ligand interactions. Receptor surface models can be constructed quickly, which allows the construction of multiple plausible models; a variable selection technique such as genetic function approximation (GFA) can then be used to suggest which receptor surface models provide the most valuable descriptors for QSAR. Advantages of this approach are shown by applying it against two previously-published and well-studied QSAR data sets. Our results indicate that the approach can model data as effectively as established 3D-QSAR techniques.