In order to select new tobacco varieties, ability of candidate lines to produce cured leaves with the desired chemical characteristics has to be assessed as early as possible in the breeding process. The high number of candidate lines renders necessary to use fast prediction tools instead of chemical analyses. For this purpose, Near Infrared Reflectance Spectroscopy (NIRS) allows estimating a number of chemical parameters, as well as more complex traits, directly from powder obtained by grinding cured tobacco leaves and this technique is now used routinely at the Bergerac Tobacco Institute. For each tobacco type, i.e. dark air-cured, Burley and flue-cured, Modified PLS (Partial Least Squares) predictive models, connecting NIR spectral data to the alkaloid, nitrogen, reducing sugar, and ash content were first developed. Models were set up from more than 1600 powder samples (crop years 1995 to 2000), including an important variability as far as spectral data and levels of chemical components are concerned. Calibration equations are based on a given population mainly at the Bergerac location. They show a high degree of performance for predicting classical chemical analyses results. Validation was performed with samples from the same crop years and external validation is also performed each year. A PLS discriminant equation allows the classification of each sample into a tobacco type according to its NIR spectrum and the predictions with the appropriate models. More recently, other criteria were investigated and predictive models are now available for chemical compounds such as NH₃, chlorine, polyphenols and smoke parameters such as tar yield from ISO 4387 smoking test and the number of revertants / mg TPM determined by the Ames test. Prospects for applications to field using a portable NIR spectrometer and for predictions from a whole tobacco leaf without grinding are also discussed.