Abstract: There are about 28 tropical cyclones per year over the Northwest Pacific Ocean and 7 tropical cyclones per year which make landfall or have significant impacts on China. Tropical cyclones lead to huge damages and economic losses. Accurate prediction of the tropical cyclone (TC) in advance would be a powerful tool in disaster preparedness and prevention.Based on the monthly mean 500 hPa geopotential height in Northern Hemisphere during 1951—2005 of 10°×10° grid points between 10°N and 85°N, monthly mean sea-surface temperature (SST) of the Pacific Ocean during 1949—2005 of 5°×5° grid points between 10°S and 50°N, monthly and annual numbers of TC over the Northwest Pacific Ocean, South China Sea, and TCs make landfalls or generate significant impacts on China or Guangdong Province during 1951—2005, the correlation coefficients between them are calculated. By analyzing the spatial distribution of the correlation coefficients between the TC frequency and the 500 hPa geopotential height and SST respectively, several key regions exist (over significant level of 5%). Those highly significant region's geographical locations are stable. The synoptic climatologically significance and the physical characteristics are investigated. Several high correlated factors are selected and combined and used to construct the binomial prediction equations to predict the TC's monthly and annual frequency in the Northwest Pacific Ocean, South China Sea, and landing ones in China and Guangdong Province respectively.The operational binomial climate prediction system is constructed and verified using the last 11-year data. The accuracy of the TC annual frequency predicted using the 500 hPa and SST is 79.6% and 77.3% respectively. The accuracy is 75.8% by using the preceding 500 hPa in November to predict the TC monthly frequency that lands in China from July to September during last 11 years.The results are as follows: Select the same sign factors to combine in the same group when two to four predictors are chosen. By analyzing high significance regions, it is found that the combined predictor method is preferable to the single sea analysis. The weighted regression ensemble analysis is superior to the general statistical methods, such as the stepwise regression analysis of principle components and empirical orthogonal function. The method has the statistical and physical significance. The prediction of the multinomial prediction equation is better than the linear prediction equation. In practice the binomial prediction equation is the best. The Binomial Climate Prediction Model can be used for many purposes and performs well.