Abstract: The initialization scheme of nonlinear balance equation is a more useful method not only in the balance model but also in the primitive equation model. But as we know that the routine method for solving the nonlinear balance equation contains the exaction of the square root, this operation consists of many operations of arithmetic in computing. Especially this calculation is to be performed at every gridpoint in every step of iteration, and to be required more iteration times. So the amounts of computations are considerably large. In this paper, a new and very efficient method for solving the nonlinear balance equation is proposed, and the difficulties mentioned above have been already removed. The experiments on medium-range numerical weather prediction are performed by using the seven-level spectral model with different initialization schemes including the nonlinear balance equation, linear balance equation, “nongeostrophic stream function”, and simply geostrophic wind. From the results, we can see that the nonlinear balance equation initialization scheme is the best one. Because of the absence of nonlinear terms in the linear balance equation initialization scheme, the forecasting results of weather system are rather smooth and weak