Abstract: Probabilistic threshold value method is often used to generate the branch for dynamic discrete event tree, which is with two deficiencies: inaccurate time of generating branch and time uncertainty for establishing of dynamic discrete event tree. This paper presents an accurate probabilistic threshold value method to overcome the deficiencies. The analysis of the dynamic cistern finds that, when the probabilistic threshold value is 0.99, with the accurate probabilistic threshold value method, the total nodes are 25 % less, the accident node are 2.7% less, and the counting time is 37.7% less that obtained by the probabilistic threshold value method. It shows that the accurate probabilistic threshold value method is with higher analytic efficiency of accident node than the probabilistic threshold value method, and the accident probability obtained by the accurate probabilistic threshold value method is 33.6% less than that obtained by the probabilistic threshold value method. In a dry out accident, it gives the operation staff longer time to deal with the accident; and the dynamic discrete event tree established by the accurate probabilistic threshold value method shows more evolution paths and accidental information, which is helpful for the operation staff to fully understand the evolution of dynamic cistern accident and suitable for the study of the system operational dynamic characteristics.