A rapid method was developed for discrimination of the geographical origins of rice with pattern recognition technique by near infrared spectrocopy (NIRS). A total of 119 geography signs product Xiangshui rice samples and 90 rice (Non-Xiangshui rice) samples produced from other places were analyzed by NIRS. After first derivative and smooth processing, principal component analysis (PCA) was used to reduce the dimensionality of the spectral data. Through the loading graph of the first three principal components, characteristic wave band (7 700~6 700, 5 700~4 300 cm-1) with max-relativity was determined. In whole wave, using agglomerative hierarchical cluster analysis and Fisher’s linear discriminant, the discrimination of Xiangshui rice and Non-Xiangshui rice was all 100%. The correct rate of specific geographical origins of Non-Xiangshui rice was 91.9% by cluster analysis and 96.7% by discriminant analysis. For analysis in the characteristic wave bands, the correct rate of discriminant by cluster analysis was higher than the analysis result through the range of the whole band. Therefore, characteristic wave band has strong representativeness. The results indicate that it is feasible to discriminate the geographical origins of rice with pattern recognition technique by NIRS, and selecting characteristic wave band is one of the validated methods to improve the precision of the discrimination mode.