The rapid discrimination of four kinds of dairy is studied by near infrared transmittance spectroscopy （NITS）. The spectra of samples are preprocessed by methods such as multiplicative scatter correction（MSC），then，they are compressed by wavelet function db3 and compression level 6. The models of discrimination which use the processed spectra as inputs are established by BP-ANN、RBF-ANN and SVM-ANN respectively. The results show that all the discrimination rates of the networks are 100% when the spectra are processed by the mixed method multiplicative scatter correction adding S-G smoothing and wavelet（MSC+S-G+WT）. Overall，the rapid and non-destructive discrimination of dairy can be realized by using NIR.