Abstract: In order to fast identify and discriminate the producing area of flue-cured tobacco leaves nondestructively, four discrimination models were established with PLS-DA algorithm on the basis of near-infrared spectra of 464 tobacco leaf samples from Sichuan, Yunnan, Chongqing and Fujian in 2008. The results showed that:1) The correlation coefficient between predicted values and measured values of classified variables of calibration set were higher than 0.94 in all the four models, which showed that the models were of good fit. 2) The discriminating accuracies of models for samples in validation set were higher than 93.0%. 3) The leaves from Sichuan, Yunnan, Chongqing and Fujian could be effectively discriminated from each other by the models.