For obtaining direct sound component effectively to analyze real characteristics of the underwater explosion source, a marking algorithm is proposed, which utilizes the local correlation peak to identify the direct sound. The evaluation function for the algorithm result is established. Based on underwater explosion experiments in the shallow water, the reasonability of this algorithm is initially verified by comparing with results through the ray theory, and further the frequency spectrum analysis is performed on the optimal direct sound interval(width about 1.8 ms) acquired by this algorithm through Hilbert-Huang Transform(HHT). The results indicate that: the sequence of the direct sound pressure level has a high autocorrelation and the corresponding relative array feature can keep stable approximately in the short range; this algorithm can bring good local correlation peaks and identify the direct sound components clearly, especially in the very short propagation range, with the correlation peaks even over 0.9; the main frequency band of the used explosion source is stable around 0-10 kHz, with the peak frequency of 1 kHz approximately, but when the propagation range is relatively longer, the corresponding spectrum profile will be influenced by the sound speed distribution significantly.