For the last 100 years, considerable effort has been oriented towards the investigation of odorous molecules from natural sources and the development of new synthetic odorants. In this search, cheminformatic tools for quantitative structure–property relationships (QSPRs) have been shown to be beneficial for the primary screening of odorant molecules. The present work attempts to identify the distinct chemical features that impart a strong smell to the molecules. Here, genetic/partial least squares (G/PLS) regression has been employed using an easily interpretable descriptor pool for generating a QSPR model to predict odour detection thresholds, using 204 diverse airborne chemicals. The model was rigorously validated using a variety of statistical parameters and different validation metrics which yielded good results. The statistical model developed provides valuable information about the contribution of structural fragments that are essential for lowering the odour threshold of the molecules, which may guide the development of new potent odorants. The interpretation of the descriptors contained in the model provides knowledge about the requirement of nucleophilicity of a molecule in relation to its binding with odorant receptors. Based on the predictive power and interpretability of the model, they might be further utilized for guiding the design and screening of new stronger odorant molecules. Copyright © 2013 John Wiley & Sons, Ltd.