Abstract: Based on the information of tropical cyclone landing at the Pearl River Gateway from the year 1949 to 1988, and with the help of a typical model of artificial neural network-B-P model,a 35－49 hour prediction model is established for the tropical cyclone landing point at the said region. And, after its application to the forecast of tropical cyclone landing point, namely, the forecast of tropical cyclone landing point at the Pearl River Gateway, the prediction model is found to have a maximum relative E-V tolerance of less than 0.7 %. The neural netowrk prediction model, as the result shows, features a stronger fault-tolerant ability as well as a faster prediction speed. Thus, it is expected to become an effective means to the forecast of tropical cyclone.