Aiming at the difficulty to identify far-region nuclear and lightning electromagnetic pulse correctly, a recognition method based on wavelet fractal technique was proposed. First, the signals of nuclear electromagnetic pulse (NEMP) and lightning electromagnetic pulse (LEMP), preprocessed by interpolation and normalization, were decomposed by wavelet packet at level two. Then, fractal dimensions of reconstructed signal from wavelet packet coefficients were calculated to form eigenvectors of NEMP and LEMP. Least squares support vector machine was chosen as the classifier and the model parameters were obtained by the five-fold cross validation. Finally, the eigenvectors were input into least squares support vector machine (LSSVM) for training and testing. The experimental results show that the combination of wavelet fractal technique and LSSVM performs well in recognition of NEMP and LEMP; the average recognition rate is more than 99%.