A reliability analysis method for complicated mechanical system is proposed which is based on Copula Bayesian network. Since the variables in complicated mechanical system are numerous and correlated to each other,the traditional analysis methods are incapable to solve the problems. However,the Copula Bayesian network takes advantage of both Copula function and Bayesian network,it can takes the correlation of variables into consideration by using Copula function,and makes probability inference of reliability at the same time. The CBN structure was transformed from BN at first,and then local t-Copulas were modeled based on the marginal distributions and correlation coefficient matrix of variables. The reliability of an jack-up gearbox was analyzed by CBN,and the correlated structure of nodes was obtained. The failure probability of gearbox was calculated by CBN,and the result showed that it was more secure and accurate than the traditional BN method.