We report a quantitative structure-activity relationships (QSAR) study using genetic function approximations to describe the activities of a polymer-carbon composite chemical vapor sensor using a novel approach to selecting a molecular descriptor set. The measured sensor responses are conductivity changes in polymer-carbon composite films upon exposure to target vapors at parts-per-million concentrations. The descriptor set combines the basic analyte descriptor set commonly used in QSAR studies with descriptors for sensing film-analyte interactions. The basic analyte descriptors are obtained using a combination of empirical and semiempirical quantitative structure-property relationships methods. The descriptors for the sensing film-analyte interactions are calculated using molecular modeling and simulation tools. A statistically validated QSAR model was developed for a training data set consisting of 17 analyte molecules. The applicability of this model was also tested by predicting sensor activities for three test analytes not considered in the training set. (c) 2006 The Electrochemical Society.