Radar full pulse repeater jamming samples through a digital radio frequency memory device and repeats according to different time sequences. Due to the absence of other modulation features within a single pulse， which is highly similar to echo signals， the feature parameters extraction is difficult， and the differences are inconspicuous in comparison with other jamming recognition. This paper proposes recognition of radar full pulse repeater jamming based on high-order spectrum features， which can suppress Gaussian white noise， contain phase information， and process non-stationary signals. It can be used to extract feature parameters， construct feature vectors， and recognize through machine learning algorithms. Finally， the algorithm is verified through a combination of practical equipment and mathematical simulation. The simulation results show that when the signal-to-noise ratio is greater than 5 dB and 25 dB， the recognition accuracy is close to 80% and 95%， respectively， which verifies the effectiveness of the algorithm and has certain value in engineering application.