Xihu longjing tea from different producing regions has different quality. Near-infrared spectroscopy, spectral pretreatment, principal component analysis and discriminant model are used to discriminate Xihu longjing tea from Longjing village, Meijiawu village and Geya village. The results show that the second derivative pretreatment method is most effective for the removal of the noise in near infrared spectra and the Bayes discriminant analysis is the best discriminant model for the tea from the above three regions. Setting the components as 5 in the Bayes model, the best original discriminant rate and the cross-validation discriminant rate are 100% and 82.35% respectively. In the cross-validation, the discriminant accuracies of the tea from Longjing village, Meijiawu village and Geya village are 80%, 83.33% and 83.33% respectively. Therefore, the model can be used to discriminate Xihu longjing tea from Longjing village, Meijiawu village and Geya village and can provide the theoretical basis for the producing region discrimination of Xihu longjing tea.