Abstract: Frequency Pulse Shaping is the measurement of the modulated phase of Continuous Phase Modulation (CPM) signals. The shape and relative length of the frequency pulse are necessary modulation parameters for the demodulation of CPM signals. In this paper, we propose a novel methodology for the recognition of CPM, which is based on circular statistics and Support Vector Machines (SVM). The sampled instantaneous frequency is treated as a set of circular random variable. The trigonometric moments are calculated and their circular statistics are used as the classification characteristics. SVM is employed to efficiently realize the classification of different pulse shaping CPM signals. Simulation results show that this algorithm realizes the recognition of arbitrary pulse shape. In this paper, we give the recognition probability among different pulses for different modulation types, including multi-h CPM signals. The classifier of SVM doesn’t depend much on the signal-to-noise conditions; therefore it has high classification performance and good robustness.