Abstract: Red light running (RLR) vehicles used to cause accidents in signal intersections. In order to solve this problem, Beijing Jiaotong University driving simulator (BJTUDS) was used to study the drivers' crash avoidance behavior based on voice warning method. Effect of voice warning information is analyzed in terms of two aspects by experiments, occurrence of collision and the corresponding harmfulness of collision. The results show that the voice information of collision avoidance warning can effectively reduce the collision likelihood as well as its severity. As a conclusion, this technology is found to be an important step for avoiding RLR crashes at intersections.