In this paper, the application of preprocessing methods in near infrared spectral data processing is presented and the particular preproeessing parameters are discussed through the example of milk. The predicted errors of the fat and protein content in milk are 0.1g/dl - 0. 3g/dl. The measurements have a good eorrelativity with the reference values. The experiment results have shown that the preprocessing method can play an important role in the optimization of models and is of great reference value to the establishment of milk models.