Abstract: GRAPES (global and regional assimilation and prediction system) is a new generation of numerical weather prediction (NWP) system of China. For such a NWP system, the design of software architecture becomes an important issue. In this paper a parallel algorithm of GRAPES is introduced. It considers not only the computing efficiency, but also the portable and sustainable development. The emphasis is put on the development of a special parallel method on the polar area in which the grid points convergence for a global version of GRAPES. Experiments are carried out on the IBM-Cluster 1600 in China Meteorological Administration (CMA). The results show that the parallel computing algorithm is correct, stable and efficient for operational implementation of GRAPES in near future.