We aim at the discrimination of varieties within a single plant species (Vitis vinifera) by means of airborne hyperspectral imagery collected using a CASI-2 sensor and supervised classification, both under constant and varying within-scene illumination conditions. Varying illumination due to atmospheric conditions (such as clouds) and shadows cause different pixels belonging to the same class to present different spectral vectors, increasing the within class variability and hindering classification. This is specially serious in precision applications such as variety discrimination in precision agriculture, which depends on subtle spectral differences. In this study, we use machine learning techniques for supervised classification, and we also analyze the variability within and among plots and within and among sites, in order to address the generalizability of the results.