In order to provide a scientific method for the geographic origin determination of millet, a portable near-infrared (NIR) spectrometer was used in combination with other methods, including principal component analysis (PCA), linear discriminant analysis (LDA) and multi-layer perceptron neural network (MLP-NN). The results showed that, for PCA model, samples from different places except Shanxi, Henan and Heilongjiang provinces clustered into different groups significantly. Millet samples from different origins could be effectively discriminated by FLDA and MLP-NN. The MLP-NN model was better than FLDA model in recognition rate. The recognition accuracy of the two models for the predication set were 92.3% and 84.6% respectively. Therefore, the NIR spectroscopy technique can be used for the geographic origin determination of the millet.