Quantitative structure–property relationship (QSPR) investigation was performed for the study of olfactory thresholds of pyrazine derivatives. Descriptors calculated from the molecular structures alone were used to represent the characteristics of the compounds. The six molecular descriptors selected by the best mutilinear regression (BMLR) in CODESSA were used as inputs for support vector machine (SVM) and radial basis function neural networks (RNFNN). The root mean squared errors (RMS) of logarithm of olfactory thresholds (p.p.m.) for the training, predicted and overall datasets were 0.5674, 0.6601 and 0.5860 for BMLR, 0.4720, 0.6861 and 0.5194 for RBFNN, and 0.5242, 0.6466 and 0.5495 for SVM, respectively. The prediction results were in agreement with the experimental values. The QSPR models provide a rapid, simple and valid way to predict the odour threshold of pyrazine derivatives. Copyright © 2009 John Wiley & Sons, Ltd.