It presents an SVMbased gait classification method to the problem of positioning accuracy in pedestrian inertial navigation because of the differences between walking and running. The algorithm divides pedestrian gaits into walking and running in real time using an SVM classifier. The original gait data was collected from an IMU installed on the foot. The establishment of training data includes coordinates transformation, FFT and dimension reduction. Tests of several walking and running data on both constant and variable speed are used to verify this method, according to the Gaussian distribution in SVM space. The results show a 98.6% successful rate of real time classification using SVM.