Data assimilation is a robust method to predict crop biophysical and biochemical parameters. However, no previous study has attempted to predict grain protein content (GPC) at a regional scale using this method. This study explored the feasibility of designing an assimilation model for wheat GPC estimation using remote sensing, a crop growth model, and a priori knowledge. The data included a field experiment and regional sampling data, and Landsat Operational Land Imager images were employed, with the CERES (Crop Environment REsource Synthesis)-Wheat model used as simulation model. To select an optimal method for data assimilation in GPC prediction, different state variable scenarios and cost function solving algorithm scenarios were compared. Additionally, to determine whether a priori information could improve GPC prediction, the collected leaf area index (LAI) and leaf N content sampling data and the range of GPC in the study region were used to constrain the data assimilation process. Furthermore, the data assimilation method was compared to the use of only the CERES-Wheat model. The results showed that GPC could be predicted by remote sensing observation, a crop growth model, and a priori knowledge at regional scale, where the use of data assimilation improved the GPC prediction compared to using only the CERES-Wheat model.