Abstract: Currently, the TC numerical prediction system at National Meteorological Center (NMC) of China is based on a global model using the initialization scheme of in serting a bogus vortex into the analysis field. Because the structure of the bog us vortex is not dynamically and physically consistent with the global model, the TC prediction system has always been run independent of the global analysis forecast cycle since it was put into operation. In fact, TC circulation is one of the important components of largescale analysis. An accurate description of initial TC structure will produce a positive effect on forecasting of large-scale circulation. In return, the large-scale circulation can steer TC's moveme nt and speed correctly at each forecast time. Obviously, the TC prediction syste m with the bogus vortex scheme is so antiquated that it cannot provide better fo recasts of TC track for operational use. Recently, a new TC numerical prediction system has been developed using a new TC initialization scheme based on the glo bal analysisforecast cycle in order to increase the skill of TC track forecasts. While the bogus vortex initialization of the old operational prediction syste m is carried out after data analysis， the TC initialization of the new system is accomplished before data analysis. It is significant that a few observations related to TC are not rejected during the quality check of the data assimilat ion, because the background field in which TC has been initialized is more consi stent with observational data. The new TC numerical prediction system for TC for ecasts over the western North Pacific was established at NMC in late 2006. To e valuate the impact of the new scheme on TC track predictions, 23 different TCs d uring the 2006 TC season were examined, which enables us to identify the case to case variability in the forecast results. It was found that the new scheme produced more accurate TC track forecass compared with the old bogus scheme. Statistical analysis shows a decrease in the average track error by 11%-23% in th e 12 to 120 h forecast time periods for the 2006 typhoon season. The new TC nume rical prediction system was applied to operational forecast parallel tests in th e 2007 typhoon season and it showed improved performance in TC track forecasts. The system was scheduled to be into operation in place of the old TC system in 2008.