Abstract: At Shuangliu International Airport the frequent, long-time low visibility often leads to the delay of scheduled flight and affects the safety of the planes. With the development of social economy and the air transport, the low visibility weather of Shuangliu Airport arouses great concern of the air control authority of the airport. Because a number of the factors can result in the low visibility weather, the factor mechanism is complex, and meanwhile the correlativity between the factors and the low visibility weather is insignificant, meteorologists are concerned most about how to forecast such weather as possible as they can. While there are huge amounts of information, the relation between the forecast object and the factors is unclear. Given this, intelligent machine learning technique is a good method to solve this sort of problems. In order to improve the forecast, Support Vector Machines (SVM), an intelligent machine learning technique that can solve the nonlinear problems is employed in the research to study the forecast method of the low visibility weather of the Shuangliu Airport. SVM method nonlinear is mapped by kernel function from lower dimensional space to higher. The higher nonlinear correlativity between the factors and the forecast object is indirectly expressed in implicit expression. Finally, the dependence of the factor and the object is depicted, and the model is set up by support vector. Hence, the model is not only related to the forecasting factor, but the kernel function as well. By using the data from Chengdu radiosound observatory and Shuangliu Airport surface observatory through 1997 to 2001, the forecast models of low visibility weather of Shuangliu Airport are built by mean SVM method with several kernel functions. Test results show the forecast model constructed with the radial base kernel function and the forecast model constructed with Laplace kernel function are better than the others, in which Threat Score are 0.287 and 0.292 separately, far above the frequency (0.155) of low visibility weather occurring at Shuangliu Airport. Test results also show, in the SVM model constructed with the radial base kernel function, the false alarm is higher and the omitted alarm is lower; in the SVM model constructed with Laplace kernel function, it is opposite. Therefore, if the accuracy of low visibility weather forecast is emphasized, the model constructed with Laplace kernel function should adopted; but if the precaution is emphasized, the model constructed in radial base function should be selected.