Winter wheat grain protein content (GPC) is an important criterion for assessing grain quality. A timely and simple GPC model is urgently required for GPC prediction ahead of maturity. The GPC model included regressional models of dry matter and N accumulation and translocation for anthesis and post-anthesis stages, and incorporated both soil nitrogen (N) supply and meterological factors based on historical as well as current season data, final GPC were calculated as the ratio of N accumulation to dry matter in grain at maturity. This study conducted six field experiments during the 2003–2006 and 2008–2011 growing seasons to establish and validate the model. A three-way factorial arrangement of N fertilization, sowing date, and cultivar was conducted using a split-plot design. Critical growth parameters were determined by field measurements, and historical seasonal meteorological data covering the growing period were collected. The normalized root mean square error (nRMSE, %), which is defined as RMSE divided by the mean of the observed value, multiplied by 100, was adopted to evaluate the model performance. The major results were as follows: (1) The prediction performance of dry matter (DM) and N accumulation (NA), and translocation during the pre-anthesis and post-anthesis periods were different; it was poorer for the former and better for the latter. However, GPC prediction was not significantly affected by the intrinsic ratio-form of the GPC prediction; (2) meteorological factors could capture the overall interannual trends of the corresponding dry matter and N sub-models in an acceptable manner; (3) nRMSE and R2 of the semi-empirical GPC model (Exp.4 and Exp. 6) were 8.91, 4.50, 0.64, and 0.46, respectively, and that of the simple linear model (Exp.4) were13.3and 0.42, respectively. The established semi-empirical model significantly improved the interannual and intra-annual prediction accuracy compared to the simple linear model.