The effect of 35 newly synthesized O-ethylmenthol (MET) derivatives on percutaneous absorption of ketoprofen was investigated in rats. In order to understand the relationship between the structure of compounds and promoting activity (structure-activity relationship), an artificial neural network (ANN) was employed. In the in vivo percutaneous absorption study, male Wistar rats, weighing 160-180 g, were used. The apparent penetration rate (Rp) was estimated based on a pharmacokinetic model with a constant rate of penetration through the skin after a lag time. As an index of the promoting activity of each compound, an enhancement factor (Ef), defined as follows, was used: Ef=Rp(with enhancer)/Rp(without enhancer). An irritation evoked on rat skin was microscopically judged at the end of the in vivo percutaneous absorption experiment and evaluated as a total irritation score (TIS). Ef and TIS were selected as output variables to determine the ANN structure. Calculated logP, molecular weight, steric energy (SE), van der Waals area, van der Waals volume, dipole moment, highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) were used as factors to determine the structural nature of cyclohexanol derivatives. Among these parameters, logP, SE and LUMO significantly affected the prediction of Ef and TIS. The predicted values of Ef and TIS coincided well with in vivo percutaneous absorption experimental values. However, results observed with a linear regression method were poor compared with the ANN approach. The contribution index of logP was approximately 50% in the prediction of Ef, suggesting that lipophilicity among physicochemical properties contributes most of the promoting activity of these compounds.