Multi-level cross entropy (MCE) method is introduced to realize the efficient and accurate estimation of thermal-hydraulic reliability (TH-R) of passive systems. The method is applicable to reliability problems with small failure probability and implicit model. Under six update strategies, the TH-R of an experimental facility is evaluated. Numerical results show that system’s failure probability can be estimated accurately with less computational cost based on MCE method, and the coefficient of variation is small. Relatively small intermediate failure probability and smoothing parameter can make the sampling center of importance sampling closer to the important area near the design point, and the evaluation accuracy is higher. Related methods and results can lead to further understanding of thermal-hydraulic uncertainty of passive system from the view of reliability, and can be used as reference for passive system’s performance evaluation and optimization.