In this study, we propose a new three-dimensional quantitative structure-activity relationship (3D-QSAR) method for selecting bioactive conformations and alignment rule simultaneously. The possible conformations of all molecules are generated by conformational analysis and they are superimposed on template conformer with possible alignment rules. The field variables are calculated as 3D descriptor of structures. Four-way partial least-squares (PLS) analysis is applied, and the conformations and alignment rule largely contributing to biological activity are selected. In order to demonstrate this method, the data set of benzodiazepine derivatives, antagonists of (CCK-B), was used as a test sample. As a result, appropriate conformers and alignment rule were selected and significant PLS model was obtained. The resulting final model could give the reasonable 3D coefficient contour maps. Moreover, external prediction was carried out by use of external data sample and its prediction was proved to be high enough.