Programs that seek to identify pharmacophoric patterns shared among flexible molecules typically rely on selecting one ligand as a template and proceed from there by applying a more or less deductive approach that discards features that are not shared by all ligands. This is not a good way to identify 3D search queries composed of some required features and some subject to a partial match constraint, however, and such complex queries are often needed to differentiate effcetively between related GPCRs. The multi-objective genetic algorithm used in the GALAHAD program generates multiple pharmacophore models, with each individual model representing a different trade-off between internal strain, feature overlap and steric overlap. In doing so, the program separates the optimization of internal and Cartesian coordinates, which removes the need for templates and allows for partial coverage models that "hit" some ligands but not all. Taken together, these attributes make GALAHAD a powerful tool for identifying discriminating pharmacophore models even when all available ligands are relatively non-specific.