The key problem of four-dimensional variational data assimilation method,which solves the constraining numerical predict equations through accompanied model,is how to establish an accompanied model.Taking the mesoscale numerical MM5 model for example,it is discussed how to set up MM5 accompanied model by making use of accompanied code technique and how to choose weighting and scaling factors in the accompanied model system.Furthermore,the conventional observations are done in the MM5 model system,and a four-dimensional variational data assimilation test is made based on observed data.The results show that this system has powerful assimilation capacity and is able to improve the accuracy of MM5 model rainfall forecast.