The potential of near-infrared spectroscopy (NIRS) was investigated for its ability to rapidly discriminate the various brands of fermented Cordyceps mycelium powder. Relationship between mycelium powder varieties and the absorbance spectra was well established with the spectra region of 12500-4000 cm-1. Spectra preprocessing was performed using 1st derivative. Principal component analysis (PCA) was adopted for the clustering analysis and re-expressing of the hyper spectral data, and then, the obtained principal components (PCs) were used as the input of back-propagation artificial neural network (BPANN) to build PCA-BPANN model for the variety discrimination. The unknown samples in prediction set were precisely identified with the correlation coefficient R of 0.9959 and root-mean-square error of prediction (RMSEP) of 0.1007, which suggests that the NIR spectroscopy, if coupled with appropriate pattern recognition method, is very promising for rapid and nondestructive discrimination of fermented Cordyceps mycelium powder.