The international market price of aromatic rice is much higher than that of non aromatic rice. There is no reliable method to authenticate the aromatic characteristic grain by grain. This situation can generate adulteration of aromatic rice batches by standard rice and leads to a lack of confidence between producers and consumers. The present study has focused on the potential of NIRS for discriminating aromatic and non aromatic rices. NIRS spectra were collected grain by grain from 13 certified aromatic cultivars and 6 certified non aromatic cultivars. Ten different grains were analysed per variety. The NIR grain analysis was performed in diffuse reflectance mode on a FOSS Nirsystem 6500 using a small ring cup equipped with a specific aluminium support. Discriminant analysis using Partial Least Squares regression (PLS2) was applied in order to sort the 190 spectra collected. The discriminant model enabled to classify rice grains with 97.4% success rate. The SECV, estimated through a "leave one out" procedure, was equal to 0.18. A validation of the model was done by predicting 20 new grains issued from 4 varieties (2 aromatics and 2 non aromatics which were represented in calibration set). This validation is a mean to evaluate the predictive power of the model and to check the hypothesis that 10 grains are sufficient to represent a variety in the database. This validation leads to a success rate of 95%: one grain only, non aromatic was misclassified as aromatic. These results are promising for establishing a method of checking rice samples, especially for detection of adulteration level of commercial batches. The principal difficulty in establishing such a database is to collect pure certified varieties. The next steps include the extension of the calibration set to other varieties and origins and to determine the optimal number of individual grains to analyse for control of one batch according to the requested limit of adulteration detection and to the model error.