Unforeseen reduction in bio-availability of drugs contribute heavily to late phase failure in drug discovery processes. P-glycoprotein, an efflux pump, that evicts a wide range of drugs is a major cause for reduction in bioavailability. Classification of potential drugs into binders and non-binders of this protein will aid greatly in weeding out the failures early in the discovery processes. The need to tap the power of computational approaches for such prediction is increasingly becoming evident, given the speed and ease with which predictions can be integrated into the discovery programs. In this paper, we report development of a prediction method to identify substrates and nonsubstrates of Pglycoprotein, based on a support vector machine algorithm. The method uses a combination of descriptors, encoding substructure types and their relative positions in the drug molecule, thus considering both the chemical nature as well as the three dimensional shape information. A novel pattern recognition method, recently reported by us has been implemented for delineating substructures. The results obtained using the hybrid approach has been compared with those available in the literature for the same data set. An improvement in prediction accuracy with most methods is seen, with an accuracy reaching over 93%.