Near infrared spectroscopy (NIRS) has been widely used to determine food chemical composition and to a lesser extent to evaluate sensory properties. Because sample preparation is relatively simple, NIRS is especially useful in situations where many samples must be analysed, such as gene-bank characterization or breeding. We aimed to assess the feasibility of using NIRS to predict aroma, flavour, mealiness, seed-coat perception, seed-coat brightness, and seed-coat roughness in common beans. Spectra of raw, undried cooked and dried cooked common bean seeds of 55 accessions were registered. Partial least squares (PLS) regression equations were developed between spectra absorbance and sensory properties scored by eleven trained panellists. Spectra registered on dried cooked samples generally yielded the best predictions. The relative ability of prediction (RAP) values were greater than 0.8 for flavour and mealiness and between 0.5 and 0.7 for seed-coat roughness and brightness. However, a suitable model to estimate the seed-coat perception was not found. These results make it possible to screen for samples that are close to the target sensory properties and thus substantially reduce the number of panel sessions needed for gene-bank evaluation or breeding. © 2013 Elsevier Ltd.