Reliability analysis is a key technology to determine the reliability of hydraulic systems and components. Bayesian networks are a common method for reliability analysis of hydraulic systems. However, in practical engineering, the uncertainty and polymorphism of the hydraulic system degrade the performance of traditional Bayesian network reliability analysis. Based on this, this paper proposes an improved reliability analysis method for Bayesian network hydraulic system based on evidence theory. Firstly, the likelihood probability and trust probability of the evidence theory are used to describe the failure probability interval of the root node, which effectively processes the uncertain information of the hydraulic system and accurately extracts the failure probability of the root node. Then, based on Bayesian inference, the fault probability of the leaf node and the sensitivity interval of the root node are obtained, which effectively describes the polymorphism of the hydraulic system. The example of hydraulic system reliability analysis shows that the present method has strong uncertain information processing ability and polymorphism analysis ability. The analysis results verify the effectiveness of the proposed method.