Structure-activity relationships (SAR) of the contraceptive progestogens for (I) oral contraceptive activity (OCA), (II) androgenic effect, and (III) binding affinity for sex hormone binding globulin (SHBG) were studied using four different methods: principal component analysis (PCA), hierarchical cluster analysis (HCA), neural networks (NN), and electronic indices method (EIM) employing descriptors calculated by the semi-empirical Austin Model I (AM1) method. An additional set of molecules was used to check the reliability of the results obtained for OCA by PCA. Using PCA, three different sets of descriptors were found to correlate with the three different biological activities, I-III, indicating that the interaction between the receptor and the progestogen must depend on the type of biological activity. The descriptors selected by PCA were also employed for SAR analysis of the contraceptive progestogens using two other methods, HCA and NN. Both HCA and NN correctly classified high activity molecules as different from low activity ones. Thus, those descriptors selected by PCA work well in the other two methods of classification. Using the sign of p, a difference of electron densities of selected molecular orbitals in a specified region in a molecule, it was possible to discriminate high activity molecules from low activity molecules in the three different types of activities studied, I-III, with one exception.20345-58