-A total of 40 milk samples, i.e., 8 milks produced in lowland (430-480 m), 16 milks produced in mid-mountain (720-860 m) and 16 milks produced in mountain (1070-1150 m) areas of the Haute-Loire department (France), sampled at key periods of animals feeding, were analysed by front-face fluorescence spectroscopy. Tryptophan, aromatic amino acids and nucleic acid (AAA+NA) and riboflavin fluorescence spectra were recorded in triplicate on different aliquots directly on milks with the excitation wavelengths set at 290 nm, 250 nm and 380 nm, respectively. The excitation spectra of vitamin A were also recorded in triplicate on different aliquots with the emission wavelength set at 410 nm. Using Factorial Discriminant Analysis, 74.1%, 81.5% and 76.9% correct classifications were obtained for the calibration data sets of tryptophan, AAA+NA and riboflavin spectra, respectively. Considering the validation data sets, correct classification of 69.2%, 76.9% and 70.4% was observed for tryptophan, AAA+NA and riboflavin spectra, respectively. In a second step, the first 5 principal components of the principal component analysis extracted from each data set (tryptophan fluorescence, AAA+NA fluorescence, vitamin A fluorescence and riboflavin fluorescence spectra) were gathered together into one matrix and analysed by factorial discriminant analysis. Correct classifications of 100 and 69.2% were observed for the calibration and the validation groups, respectively. Milks from the lowlands were well discriminated from the others. It is shown that fluorescence spectra of milks retained information related to the geographic origin. However, the size of the data set should be increased in order to improve the robustness of the algorithm. AAA+AN et de la riboflavine. En ce qui concerne les jeux de validation, des bonnes classifications se montant à 69,2 %, 76,9 % et 70,4 % ont été observées respectivement pour les spectres des tryptophanes, des AAA+AN et de la riboflavine. Dans une deuxième étape, les 5 premières composantes principales des analyses en composantes principales réalisées sur chaque jeu de données (spectres de fluorescence des tryptophanes des protéines, des AAA+AN, de la vitamine A et de la riboflavine) ont été collées les unes aux autres et l'analyse factorielle discriminante a été appliquée à cette matrice. Les résultats indiquent que 100 % et 69,2 % de bonnes classifications ont été observés respectivement pour les échantillons principaux et de validation. Les laits de plaine sont bien discriminés des laits de moyenne montagne et de montagne. Les résultats montrent que les spectres de fluorescence des laits contiennent des informations relatives à leur origine géographique. Cependant la taille des jeux de données devrait être augmentée afin d'améliorer la robustesse de l'algorithme.lait / origine géographique / fluorescence intrinsèque / chimiométrie