In this paper we choose the SVM (support vector machine) method for forecasting 2min average wind speed four times daily (00, 06, 12 and 18 UTC) when it is affected by landfalling tropical cyclones. First we select related factors based on the intensity of the tropical cyclones during 2002-2007, the landform and the environment element variables at lowlevel and upperlevel around the station. And then we establish numerical weather prediction models. The standard deviation of the wind speed fitting error in Model 4 is 1.591 m·s-1. By testing with 8 landfalling tropical cyclones in 2008 as independent samples, the difference of the actual average absolute wind speed from the forecast one is 1.75 m·s-1, and the standard deviation is 2.367 m·s-1. The precision of wind speed forecast in 48 h can be better when the SVM method is used under conditions of selecting appropriate forecast factors and sample truncation.