With the development of adaptive optics (AO) technology in laser field, a variety of improvement measures based on software monitoring and hardware protection have been added to the classical AO system to ensure stable and continuous light output of laser AO system. Facing the reliability challenge brought by the increase of structural complexity, how to build a system failure model to evaluate the reliability of laser AO system has become an important part of the development of laser AO system. In this paper, a dynamic fault tree (DFT) method is proposed to evaluate the reliability of laser AO system, and the dynamic fault tree is established according to the dynamic relationship between the equipment. The bottom event failure rate is estimated by combining the manufacturer information, fatigue life test and historical data. The reliability parameters of DFT are obtained by using binary decision graph and Markov model. Using DFT to analyse the reliable running time of the AO system with the improvement measures, the result shows more than ten times improvement relative to the basic fault tree. In the actual system joint commissioning, no self-induced failure occurred during the expected reliable running time, which is consistent with the DFT estimate. It is proved that the reliability evaluation of laser AO system with improvement measures is more accurate by using DFT method.