The human ATP-binding cassette (ABC) transporter ABCG2 plays critical roles in cellular protection against xenobiotics as well as in multidrug resistance in cancer chemotherapy. We have developed the quantitative structure-activity relationship (QSAR) to analyze ABCG2-drug interactions. For this purpose, plasma membrane vesicles prepared from ABCG2-expressing Sf9 cells were used as a model system to measure the ATP-dependent transport of [³H]methotrexate.  Based on their inhibition profiles with structurally diverse test compounds, we performed QSAR analysis using chemical fragmentation codes deduced from the structures of the test compounds.  Multiple linear regression analysis delineated a relationship between the structural components and the extent of ABCG2 inhibition, allowing us to identify one set of structure-specific chemical fragmentation codes having close correlation with the inhibition of ABCG2 transport activity. Based on the QSAR analysis data, we analyzed the potential of gefitinib to inhibit ABCG2. Furthermore, we tested camptothecin (CPT) analogues to gain insight into their interactions with ABCG2. CPT analogues with a hydroxyl group at position 10 or 11 of the A ring are recognized by ABCG2 and are thereby effectively extruded from cancer cells. Electrostatic potential iso-surfaces generated by ab initio MO calculations using the restricted Hartree-Fock method have revealed that a negative potential localized at position 10 or 11 in the A ring is important for recognition by ABCG2. MO calculation-based neural network QSAR analysis would provide a useful approach to understand  the substrate specificity of ABCG2 and also to design new anticancer drugs that circumvent ABCG2-associated drug resistance.