A method is put forward to recognize the fog based on the support vect or machine, according to the satellite remote sensing data. Firstly, the probabi lity statistics method is used to roughly judge the fog, according to the freque ncy of the fog areas appearing at different channels of FY1D satellite; second ly, based on the former judgment, the support vector machine is applied to judge the fog carefully; lastly, erosion and dilation techniques are used to optimize the result of the second procedure. From September to December in 2006, 65 over passes of FY1D satellite data including fog areas are analyzed, and the judged fog areas are found to correspond well to the experts' experience. And the resu lt shows that the combination of 1, 2, 4, 6, 7 and 10 channels is the best of ju dgment. The 5fold crossvalidation is 89.9849% and the TS score is74.04%. This method is also used to recognize the fog during other time, and fou nd that this method is excellent.