A novel 3D QSAR approach, comparative spectra analysis (CoSA), in which molecular spectra are used as three-dimensional molecular descriptors for the prediction of biological activities, is presented and discussed. To this purpose, experimentally determined 1H NMR, mass, and IR spectra, as well as simulated IR and 13C NMR spectra, for a set of 45 diverse progestagens are converted by a program, SpecMat, into matrixes, which are subsequently employed in a multivariate regression analysis (PLS). The results are compared with those resulting from a comparative molecular field analysis (CoMFA). When used individually, spectral descriptors yield better correlations and predictions than molecular field descriptors. A combination of spectral descriptors with other descriptors, either spectral or molecular field in nature, leads in most cases to models that are statistically superior to the ones obtained by their corresponding individual spectral or molecular field descriptors.