Abstract: 【Objective】To analyze the genetic variability of protein content in rice hybrid progenies, the feasibility of building calibration model to estimate crude protein content in intact rice kernel by near-infrared reflectance spectroscopy (NIRS) was studied. It is beneficial to provide the parameter and improve nutritional quality in grain breeding program. 【Method】Using 191 rice samples with a broad range of crude protein content, 5.90%-14.50%, as materials, the chemometrical method of partial least square regression was used to establish the calibration model. 【Result】 The optimal model was developed by the spectral data pretreatment of the first derivative + vector normalization in 11 998.9 cm-1-5 449.8 cm-1 and 4 601.3 cm-1-4 246.5 cm-1, by analyzing different spectral data pretreatment and light frequency ranges. This model's determination coefficient and relative standard deviation (RSD) were 0.9886 and 0.021, respectively. The model showed significant correlation and lower error between near-infrared value and true value, and had lower root of mean square errors than 0.4. The crude protein contents of another 205 seed samples (nondestructive brown rice) from the F2 populations of 20 crosses were estimated by the calibration model. The Mahalanobis distance of 203 individuals in F2 populations predicted by the model were lower than 0.3, and showed higher calibration precision. The results showed those populations had wide variation among individuals in crude protein content. In general, the great majority was in the middle of both counterpart parent, but some of them were found higher or lower than their parents in crude protein content, and the highest value reached 15.3%. 【Conclusion】Good calibration equation was successfully developed for protein content, the equation show satisfactory determination coefficients. The samples were scanned by NIRS and analyzed the protein content in parents and hybrid progenies. Finally, a probably effective way to improve protein content of rice was proposed: Firstly，the breeder makes some crosses between promising rice varieties with high yield and good quality and rice resource germplasm with high protein content, and secondly, the NIRS is used to evaluate protein content of the nondestructive brown rice in their early progeny population. This NIRS-assisted-selection could be a very efficient method to improve protein content in rice breeding programs.

Key words: Brown rice, NIRS (near-infrared reflectance spectroscopy), Partial least-squares regression (PLS), Calibration model, Protein content, Rice breeding