A quantitative structure-activity relationship model has been developed for three important classes of psychotropic drugs including phenothiazines, antidepressants and anxiolitics. Toxicity was expressed as the oral LD(50) values for mice. Acute toxicity of drugs correlated with a proposed descriptor based on the oxidation mechanism of bioactivation. The new descriptor was named hydrogen charge (HC) descriptor, and it is the sum of the HCs of amine moiety, carbons in the alpha-position of the heteroatoms, carbonyl groups and unsaturated bonds. The results showed that the model with the descriptor of ratio of the HC over volume of the molecules, named the effective hydrogen charge (EHC) were superior to the model with the HC descriptor. Multiple linear regression was derived for the three categories of drugs separately predicting the LD(50) values. The linear models were found to describe the system with statistical parameters of RMSECV and RMSEP as follows: 41, 49 for phenothiazines, 111, 190 for antidepressants and 151, 276 for anxiolitics. The performance of the constructed models was examined using RMSECV (leave-one-out and leave-group-out), standard error, coefficient of determination (R(2)), R(2) cross-validated (q(2)), the average chance correlation and residual error plots. Finally, the LD(50) for a variety of compounds, with unknown experimental values, were predicted.