Numerical simulation of groundwater is often affected by the uncertainty of model generalization, observation errors and so on. Therefore, it is very necessary to quantitatively analyze these uncertain factors. In this study, The DREAM (Differential Evolution Adaptive Metropolis) algorithm is combined with the MODFLOW and applied to quantitatively analyze the uncertainty of the numerical simulation results of groundwater. For an ideal groundwater simulation model, the uncertainty of the simulation results which affected by the uncertainty of model structure generalization and the head observation errors is systematically analyzed. The study results show that the uncertainty of model structure generalization and the head observation errors cause the uncertainty of the groundwater simulation simultaneously. However, the uncertainty of model structure generalization is the controlling factor. When the model structure is correct, the uncertainty of model parameters and simulation results are small, and increase with the increasing of the observation errors. When the model structure is incorrect, the uncertainty of model parameters and simulation results are significantly increased, and mainly affected by the uncertainty of model structure generalization. Furthermore, under the same observation errors, the model structure generalization is more close to the real hydrogeological condition, the uncertainty of model parameters and simulation results is smaller.