Abstract: Direct sequence spread spectrum (DSSS) signal is difficult to detect due to its low power spectral density characteristics. Aiming at the problem that the performance of traditional auto-correlation detection method for DSSS signal degrades sharply under the condition of low signal-to-noise ratio (SNR), according to the analysis of autocorrelation characteristics of DSSS signal, a detection method based on generalized cross-correlation (GCC) estimation is proposed. Firstly, the received signal is segmented and the generalized cross-correlation estimation for adjacent segmented signals is performed successively, then the estimated results are processed by non-coherent accumulation of the second-order moment to extract cross-correlation peaks as detection statistics, which are compared with a certain threshold to determine whether a signal exists or not. The comparison between the performances of the improved autocorrelation detection method and the proposed method in this paper is made by computer simulation, and the result shows that the SNR tolerance of the proposed method is reduced by 5 dB compared with the improved auto-correlation method, and the proposed method can detect DSSS signals under low SNR condition. The processing result of pool test data also verifies the practicability of the proposed method.