The structures and molecular properties of 95 aromatic and heteroaromatic ligands previously tested as reversible inhibitors of chymotrypsin catalysis have been calculated using AMl. The properties obtained have been used as input for multiple linear regression analysis and as descriptors for a back-propagation neural network to predict the binding affinity of alpha-chymotrypsin inhibitors. Using polarizability, molecular shape, electrostatic similarity, dipole moment, ClogP, and the diagonalized quadrupole moments of the ligands, correlation coefficients between calculated and experimental affinities of 0.96 for the training set and 0.89 for the test set were obtained using a neural network. The performance of the multiple linear regression was significantly worse, although useful QSARs were also obtained.