The purpose of this study was to develop a quantitative structure-activity relationship (QSAR) for the prediction of the apparent volume of distribution (Vd) in man for a heterogeneous series of drugs. The relationship of many computed, and some experimental, structural descriptors with Vd, and the Vd corrected for protein binding (unbound Vd), was investigated. Models were constructed using stepwise regression analysis for all the 70 drugs in the dataset, as well as for acidic drugs and basic drugs separately. The predictive power of the models was assessed using half the chemicals as a test set, and revealed that the models for Vd yielded lower prediction errors than those constructed for the unbound Vd (mean fold error of 2.01 for Vd compared with 2.28 for unbound Vd). Moreover, the separation of the compounds into acids and bases did not reduce the prediction error significantly.