A novel QSAR study of benzamidines complement-inhibitory activity and benzene derivatives acute toxicity is reported and a new efficient method for selecting descriptors is used. Complement-inhibitory activity QSAR models of benzamidines contain from one to five descriptors. The best, according to fitted and cross-validated statistical parameters, is shown to be the five-descriptor model. Models with a higher number of indices did not improve over the five-descriptor model. The benzene derivatives structure-toxicity models involve up to seven linear descriptors. Multiregression models, containing up to ten nonlinear descriptors, are also reported for the sake of comparison with previously obtained additivity models. Comparison with benzamidine complement-inhibitory activity models and with benzene derivatives toxicity models from the literature favors our novel approach.