In vitro human skin permeability coefficient data collected by Flynn (1990) have been analysed using multiple regression analysis. An improved model for the prediction of permeability coefficients has been derived by the inclusion of the melting point as an independent variable in addition to the octanol-water partition coefficient (as logP) and molecular volume. (The solubility of chemicals in water is related to logP and melting point through an algorithm; Suzuki, Journal of Computer-Aided Molecular Design 1991, 5, 149-166.) Examination of the dataset using principal components analysis confirmed that it could be divided into three distinct groups of chemicals-steroids, other pharmacologically active molecules and small molecules. Regression analysis of the individual groups revealed a very high level of correlation with the model for the steroids and small molecules, but a poor correlation for the pharmacologically active molecules. A sub-set of hydrocortisone derivatives within the steroid group had measured permeability coefficients which were around 1.5 orders of magnitude greater than values predicted from the small molecule model. By grouping together the small molecules and the steroids (excluding the hydrocortisone set)-60 molecules-a quantitative structure-activity relationship (QSAR) has been derived for their permeability coefficients which is dependent on logP, molecular volume and melting point, and which explains more than 90% of the variability in the data. Although yet to be tested experimentally, this QSAR is expected to give accurate predictions of in vitro human skin permeability coefficients within the constraints of the dataset.