The results are presented from an empirical study on the distribution of dilemma zones for different groups of drivers at signalized intersections. With a specially designed video-based system, this study conducted extensive field observations of the responses of 1,123 drivers to a yellow phase at six intersections with high accident frequency, including all critical data such as the speed evolution during the yellow phase, the acceleration and deceleration rates, and the approximate reaction time to an encountered yellow phase. These empirical results revealed that the dilemma zone is dynamic in nature, with its location varying with the driving population, and that the recommended and commonly used practice of extending the yellow phase duration may not eliminate all the dilemma zones. Two types of strategies that can effectively eliminate the dynamic dilemma zones are also designed to improve intersection safety.