Abstract: The characteristics of wheat protein content has high heritability, so fine-quality breeding can be achieved by selecting the high-protein wheat seed. Combined with chemometric methods' hyperspectral imaging technique was used to build the average model to achieve fast prediction of single wheat seed protein content. In the experiment, 47 unit wheat seed samples' hyperspectral images were collected by GaiaChem-NIR system, and the average spectra was obtained by image process methods. Then, synergy interval partial least squares was applied to select the characteristic spectral regions to optimize the prediction model of wheat seed protein content. The optimal models' determination coefficient is 0.94, the root mean square error of prediction is 0.28%, and the residual predictive deviation(RPD) is 3.30. Finally, the average model was applied to predict the protein content of each pixes of single wheat seed, and calculated the average as the single wheat grain protein content. The experimental results showed that different wheat grain's protein content value predicted by the optimal model existed difference. Ueanwhile, the prediction values varied around the average protein content of its sample, which reflected that the average model is accurate and feasible to predict single wheat grain's protein content. Therefore, the studied method provides a new way to select the high-protein wheat seed in the process of breeding, which can promote the development of wheat fine-quality breeding.