Abstract: The blind detection of burst signal is extremely important in the non co-operative communication. The key technology of the blind detection for the burst signal is the features construction and precise start point capture. A new blind detection algorithm based on the cyclic spectrum is proposed in the paper. Firstly it provides the relative variance of cycle frequency profiles data derived from the cyclic spectrum as the detection feature, which can effectively depress the interference caused by the strong noise. Secondly, the variance of the detection feature is used to estimate the existence of the useful signal, and it has a good correct detection ratio. Then, the searching algorithm based on the peak value of detection feature is designed to catch the start point and the end point of proruption signal accurately, and it can avoid many problems caused by the setting of threshold, which is used to distinguish signal from noise. Simulation results show that, the algorithm proposed in the paper has a good performance, it can depress the interference caused by the strong noise, has a good detection ratio of the useful signal, and under the strict error control circumstance, it can catch the start point and the end point of proruption signal more accurately than the algorithm based on the ratio of the energy of two adjacent segment signals.