In most multi-function phased array radar applications,multiple missions, including airspace searching and target tracking,are usually performed simultaneously by the digital beam-forming technique and the time dividing method. This paper presents a novel method to classify pulses of different missions from an interleaved pulse sequence emitted by the same radar, which is significant in radar electronic reconnaissance and electronic support measure. Firstly, two hypotheses, i.e., pulse relativity within the same mission and pulse independence among different missions, are proposed by analyzing the antenna pattern and the beam scheduling method of the phased array radar. Based on the above two hypotheses, an optimal model for pulse classification is exploited with pulse amplitude series, where the absolute-value sum of second order difference is taken as the optimal kernel to measure sequence smooth continuity. Finally, several pieces of sequences under different numbers of missions and tracking data rates are simulated for algorithm verification. The simulation results show that the long data length and the high data rate will increase classification efficiency due to the validity of the two hypotheses in sufficient pulse amplitude sequence. In most multi-function phased array radar applications, multiple missions, including airspace searching and target tracking, are usually performed simultaneously by the digital beam-forming technique and the time dividing method. This paper presents a novel method to classify pulses of different missions from第一, 两 hypotheses, ie, pulse relativity within the same mission and pulse independence among different missions, are proposed by analyzing the antenna pattern and the beam scheduling method of the phased array radar. Based on the above two hypotheses, an optimal model for pulse classification is exploited with pulse amplitude series, where the absolute-value sum of second order difference is taken as the optimal kernel to measure sequence smooth continuity. Finally, several pieces of sequences under different numbers of missions and tracking data rates are simulated for algorithm verification. The simulation results show that the long data length and the high data rate will increase classification efficiency due to the validity of the two hypotheses in sufficient pulse amplitude sequence.