The continuing decline in coffee prices and the deterioration of coffee plantations has led Central American producers to reconsider their production, particularly their choice of arabica cultivars, with a view to winning new markets (terroirs, origins, etc.). In order to guarantee certification for these coffees and a quality bonus, reliable and rapid control methods need to be developed. Near infrared reflectance spectroscopy was investigated as a means of discriminating between 3 cultivars (CR95, Caturra and Sarchimor) distributed in Costa Rica. This study involved 36 samples obtained from an experimental plan crossing 3 factors: cultivar, harvest date and degree of ripeness. The samples, which came from the same plantation, underwent the same post-harvest treatment to obtain green coffee. The water, caffeine, chlorogenic acid, trigonelline, fat and sucrose contents of these 36 samples were determined in the laboratory and the infrared spectra were acquired in diffuse reflectance using a Foss 6500 monochromator. An analysis of variance was carried out to test how the different factors affected the contents of the six constituents, and a factorial discriminant analysis based on these biochemical variables made it possible to class the samples by cultivar, with a rate of 94.4%. A cluster analysis using Ward's criteria was applied to the matrix of Euclid distances extracted from the spectral data, making it possible to detect a grouping of the samples preferentially by cultivar. Different classification methods based on spectral data were tested, such as partial least squares, discriminant analysis based on principal components and stepwise discriminant analysis. The discriminant model developed from the principal components led to 100% successful classification of the samples according to the cultivar. The performance of the model was tested, with a 100% correct classification rate, on a set of independent samples. These results are promising for introducing a way