Snow lotus samples from four different geographical origins (Qinghai, Tibet, Yunnan and Sinkiang) were studied. K-nearest neighbors (KNN）algorithm was applied to build discriminating model as a pattern recognition method. The parameter K of the KNN model and the number of principal component factors (PCs) were optimized. The spectra were preprocessed by four different spectral preprocessing methods of standard normal variate (SNV), multiplieative scatter correction(MSC), first derivative and second derivative, and their effects on results of KNN models were compared. Experimental results showed that the optimal model was obtained with PCs 5 and K=3 after SNV spectral preprocessing, and the discriminating rates were all 100% in cross-validation and prediction. The work demonstrated that NIR spectroscopy technique with KNN method could be successfully applied to discriminate snow lotus from different geographical origins. 