Reliability as an important indicator of the normal work of smart energy meter is the focus of power quality research at this stage of the problem, this paper aims at the importance of intelligent energy meter reliability as the starting point to discuss the reliability of the energy meter research process to Bayesian The traditional method of Bayesian network"s reliability prediction bi-direction reasoning is analyzed and a new bucket elimination method is introduced to solve the Bayesian network in the process of solving the reliability problem. At the same time, based on the model theory of the fault tree, Network to simplify the calculation, and thus quickly get the reliability of the expected results. Finally, the method proposed in this paper is applied to the case of reliability prediction of electric energy meters. The results of the calculation meet the requirement of GJB / Z299C-2006 standard in the error range, which proves that the method can simplify the prediction process and accurately predict the reliability at the same time. result.