The icing of overhead transmission lines is a serious problem for the safety of power grid. A method based on support vector machine (SVM) is presented here to predict the icing load of power transmission line, which utilises the history meteorology data from the icing monitoring system, and takes account of time effects of meteorology factors. According to the results of simulation, the prediction of icing load on the power lines in Northeast of Yunnan province is close to the actual value and the ability of generalisation is better than the model based on BPNN.