Abstract: Unsystematic error is one of main sources of model simulation error, which is mainly induced by initial field error and model defect. Global and Regional Assimilation and Prediction System (GRAPES) global model forecast data and final analysis data made by National Centers for Environmental Prediction (NCEP) during January, April, July, October in 2014 are chosen to be compared and analyzed. In terms of temporal variation, conclusion could be made that peaks of unsystematic errors in both north and south hemispheres occur in their respective winters, and errors show periodical changes. With the increase of forecasting time, the model unsystematic mean square error along with geopotential height field increases over time, first in an exponential function way, and then linear growth. In addition, linear growth in temperature field and zonal wind field are discovered. It shows in the consequence that the large value of model unsystematic mean square error center in mid-latitude and distribute like zonal banded. Large value regions basically do not change along with forecast time. In zonal average geopotential height field and zonal wind field, large value regions are found in tropopause, whereas it is found in boundary layer in temperature field. The cause is that the parameterization scheme does not fully describe differences of these two stratifications in physical process and dynamic framework. It is worth mentioning that the error of the increase in the height of the temperature field at middle and upper levels of the troposphere decreasing. After fitting the error-time line, the error of the initial field, the upper limit of the prediction and the proportion error taken up by the initial field in the south hemisphere are all higher than that in the north hemisphere. Besides, it is found that the initial field gradually decreases in proportion with height increasing. It shows in the above results that the precision of GRAPES_GFS 2.0 for the simulation of mid-latitudes of the south hemisphere and the entire troposphere should be further improved. In order to discover internal physical mechanism of model forecast's defect, and correct error target, it is necessary to research the feature of unsystematic error of GRAPES global model, and analyze the spatial-temporal evolution of the error.