The computation consumption of Bayesian network increases exponentially with the nodes number and that limits the application of large－scale Bayesian network. In order to relieve this situation, this paper makes use of the independence condition to decompose the whole Bayesian network according to the characteristics of safety analysis, compresses the items number in reasoning process, proposes the algorithms for top event probability and cut sets, simultaneously analyses the computational complexity. Compared with BDD－based FTA algorithm, the proposed algorithms showed a lower memory demand and a higher speed performance when meeting the need of safety engineering.