Abstract: The Sixty-three representative Pollock fish samples were collected, in order to establish a near-infrared quantitative analysis model by fitting the pollock fish near infrared diffuse reflectance spectra of water, fat and protein content. The fitting was conducted by Partial Least Squares (PLA) , verified by the internal cross-validation method, compared and analyzed with different pretreatment methods and main factor of the spectrum. The optimum pretreatment method and the main factor were obtained. The correlation coefficients of models of water, fat and protein were 0. 9342, 0. 8125 and 0. 9113, with RPD of 3. 4, 2. 3 and 3. 1 respectively. Results showed that the near-infrared models of water, fat and protein were of high accuracy and stability, which could determine the content of water, fat and protein rapidly without destruction.