Abstract: In this paper, a number of numerical experiments are carried out with area meso scale four-dimensional variational data assimilation system. In those experiments, initial condition, modelerror and boundary condition are selected as control variable to adjust. The results of data assimilation and following prediction are tested. The main propose of those experiments is to find out the actions of modelerror and boundary condition in data assimilation, and to select the method for those three as control variable at same time. The results show that modelerror and boundary condition have large affection to forecasterror except initial condition in area mesoscale model. The experiments use a method to correct modelerror, and the results show this correction makes cost function converge quickly in data assimilation and has persistence in the following prediction. The adjust ment to boundary condition makes cost function converge quickly. The results also show that initial condition, modelerror, and boundary condition should select as control variable at same time in data assimilation. Otherwise, it will transpose one error to another error because the model has noability to disting uish where forecast error really comes from, and makes cost function converge quickly but the following forecast has no improvement.