Abstract: Weather forecast is one major application area of high performance computing technology. The running of meteorological numerical models demands strong high performance computing resource support to ensure the timeliness of numerical weather prediction systems. However, high performance computing resources and supporting capabilities are characterized by geographically contagious distribution in CMA. Local meteorological bureaus are well behind national institutions, whether in the possession of HPC resources or application development capabilities. High performance computing in meteorological field has some typical features in accordance with the requirements for grid computing, such as computational intensiveness, distributed and cooperative mass data access. Regarding the requirements of resource integration, sharing and management by local and national institutions in CMA, a design scheme of nationwide meteorological computational Grid is proposed. Grid technology is used to form abstract virtual resources on heterogeneous computing resources in meteorological department, so as to shield the heterogeneity of the underlying physical systems. Through orderly management and collaborative computing, the service platform implements effective aggregation and comprehensive utilization of resources. The design scheme employs a 3 level layout of national, regional and provincial nodes, constituting a distributed, tightly coupled network computing sharing system. The nodes are interconnected by WAN based meteorological broadband network. Upon the resource aggregation platform, function modules are intercalated for resource management, application services and user interfaces. With key technologies like UNICORE, function modules are developed and implemented. 6 geographically distributed nodes are established. UNICORE gateway services are deployed onto the meteorological broadband network, interacting with one another via grid communication protocols. 9 heterogeneous high performance computers in different places have been integrated and make up a meteorological computing resource pool. Two types of sharing services are provided, grid middleware and customized operations, on the nationwide meteorological computational Grid. By means of customized operations, three model application systems are set up. Since establishment and operational running, the meteorological computational Grid disseminates numerical weather prediction products to users in remote or resource poor areas, and thus provides strong support for disastrous weather prediction services and important events meteorological assurance, which plays an important role in local disaster prevention and mitigation efforts.