Abstract: A near infrared analysis model of protein contents in nutritious infant rice flour was established by using near infrared spectroscopy and partial least square method. The spectral stability of the sample with different sample thickness 2, 3, 4, 5, 6, 7, 8 mm were compared. The standard deviation of the sample spectrum under different wave number determined that 5 mm was the best loading thickness. The effects of variable standardization, first order derivative, multivariate scattering correction and 9-point smoothing preprocessing method on the model establishment were compared. According to the correlation coefficient and root mean square error of the model, the best pretreatment methods were variable standardization, first derivative and 9-point smoothing. The near infrared quantitative analysis model of protein contents in infant nutritious rice flour was established. The calibration correlation coefficients of calibration set and verification set were 0.9856, 0.9841, and the root mean square error were 0.2431%, 0.2456%, respectively. The established NIR quantitative analysis model was verified by external samples. The accuracy of prediction was 96%. The results showed that the near infrared detection method was consistent with the national standard method, and it was fast and didn't need sample pretreatment. It would have wide application value in the analysis of protein content of infant nutritious rice flour.