The in vitro antiviral activity of 28 nucleosides against the parainfluenza virus type 3 has been analyzed by using a novel computer aided receptor modeling procedure. The method involves an extensive modification of our earlier work (Ghose, A. K.; Crippen, G. M. J. Med. Chem. 1985, 28, 333). It presents a more straightforward algorithm for the steps that suffered from subjectivity in the earlier method. The method first determines the possible low-energy conformations of the nucleosides, and assigns a priority value for each conformation of each molecule. It then performs the following steps repeatedly, until it finds an acceptable solution. Starting from the conformation of highest priority, the various energetically allowed conformations of the other molecules are superimposed on it. On the basis of the physicochemical property matching (or overlapping), the best superposition is determined. The superimposed molecules are dissected into a minimum number of parts and the local physicochemical properties at different regions are correlated with their binding data (antiviral activity). A modified version of distance geometry has been used for geometric comparison of the structure of the molecules. On the basis of the virus rating (VR) of 28 ribonucleosides, this procedure hypothesized the minimum-energy conformation of 6-(methylthio)-9-beta-D-ribofuranosylpurine as a reference conformation and used three physicochemical properties, namely hydrophobicity, molar refractivity, and formal charge density for property matching. The binding-site cavity was divided into seven regions or pockets to differentiate the nature of interaction quantitatively. The model suggests that the 2- and 3-positions of the purine ring and the corresponding atoms of the other rings get some steric repulsion, and nucleosides having a single five-membered heterocyclic ring will better fit this virus. The methylthio group gets a strong attraction from dispersive interaction. Both hydrophilic and dispersive groups are attractive here. Although our calculation supports the previously suggested active conformation of ribavirin, it shows that it is not the global minimum-energy conformation. The difference lies in the orientation of the amide group. The calculated viral rating from this model showed a correlation coefficient of 0.971 with the observed values, and the explained variance and the standard deviation of the fit were 0.880 and 0.125, respectively.