The thickness of grain pericarp, the outer layer of the kernel, is an important breeding criterion for sorghum. This cereal is mainly used through traditional processing in family-based food systems in many regions of the world. We investigated in this study how pericarp thickness could be predicted by Near Infrared Reflectance Spectroscopy (NIRS), a fast and non-destructive measurement method that is commonly used to measure physico-chemical parameters of sorghum grains, and how this trait also influences the prediction of those parameters. We showed that, using a classification approach, it was possible to discriminate thick from thin pericarp whole grain samples with a good accuracy and that the proportion of thin and thick grains in mixed samples could also be predicted. In addition, pericarp thickness had a significant effect on the calibration performance for other grain parameters indicating that the pericarp can distort spectral information of whole grain samples. As a practical consequence, we suggest to develop separate whole grain calibration models for thin and thick pericarp samples, combined with a two-steps prediction approach to improve the accuracy of whole grain NIRS calibrations for grain quality parameters in sorghum.