Near infrared spectroscopy (NIRS), coupled with principal component analysis and wavelength selection techniques, has been used to develop a robust and reliable reduced-spectrum classifi- cation model for determining the geographical origins of Nanfeng mandarins. The application of the changeable size moving window principal component analysis (CSMWPCA) provided a notably improved classification model, with correct classification rates of 92.00%, 100.00%, 90.00%, 100.00%, 100.00%, 100.00% and 100.00% for Fujian, Guangxi, Hunan, Baishe, Baofeng, Qiawan, Sanxi samples, respectively, as well as, a total classification rate of 97.52% in the wavelength range from 1007 to 1296 nm. To test and apply the proposed method, the procedure was applied to the analysis of 59 samples in an independent test set. Good identification results (correct rate of 96.61%) were also received. The improvement achieved by the application of CSMWPCA method was particularly remarkable when taking the low complexities of the final model (290 variables) into account. The results of the study showed the great potential of NIRS as a fast, nondestructive and environmentally acceptable method for the rapid and reliable determination for geographical classification of Nanfeng mandarins.