Abstract: In this paper the performances of the conventional TDOA acoustic source localization method in reverberant environments are studied when the sound source is an impulsive signal. Based on the Generalized Cross-Correlation and Time Difference of Arrival estimation method (GCC-TDOA), three measures are put into use to enhance the robustness of the original method. The overlapping windowing for data batch processing is first used, and then a series of constraints on the cross-correlation function are defined to exclude the abnormal time-delay estimates. The energy-weighted voting is used as a post processing step to further increase the stability. As a result, the direction estimation performance is significantly improved in reverberation. This algorithm can be used to localize the unstationary signals like gunfire, explosions and so on.