The purpose of regional simulation is to predict the variability of yield at both spatial and temporal scale, by using very limited geographical data. However, there are some uncertainties which related to regional simulation, e.g. model complexity, imperfect data, regional calibration and validation etc. This study demonstrated how the regional calibration works by using the CERES-Rice model. Firstly, the model was calibrated and validated at site specific scale; then two regional calibration approaches, average approach and RMSE approach were employed to identify a representative cultivar for each sub-AEZ (Agro-Ecological Zone). The results shows: the regional simulation can simulate the yield variability spatially and temporally, RMSE approach is better than average one. Further endeavor was needed to decrease the uncertainties of regional simulation.