A new method for the discrimination of varieties of near acidophilous milk by means of near infrared spectroscopy (NIRS) was developed. Firstly, through the principal component analysis (PCA) of spectroscopic curves of 5 typical kinds of acidophilous milk, the clustering of acidophilous milk varieties was processed. The analysis results showed that the cumulate reliabilities of PC1 and PC2 (the first two principal components) reached 98.96%, and the cumulate reliabilities of PC1 to PC7 (the first seven principal components) were 99.97%. Secondly, a discrimination model of artificial neural network (ANN-BP) was set up. The first seven principal components of the samples were applied as ANN-BP inputs, and the values of type of acidophilous milk were applied as outputs, then the three layer ANN-BP model was build. In this model, every variety of acidophilous milk includes 27 samples, the total number of samples is 135, and the rest 25 samples were used as prediction set. Calculation results showed that the distinguishing rate of the five acidophilous milk varieties was 100%. This model is reliable and practicable. So a new approach to the rapid and lossless discrimination of varieties of acidophilous milk was put forward.