Abstract: Bayesian network has a powerful processing capacity and the capacity in self-learning and upgrading on the uncertain problems. Application of Bayesian network software can improve the effectiveness of risk prediction related to the network. In a urban gas pipeline case, Bayesian network-based model for the failure probability of urban gas pipeline is built to calculate the failure probability of polymorphism top event and structural importance containing failure factors by HUGIN and MSBNX software tools. By means of BN's reasoning ability, singlefactor and two-factor corrections on the natural destruction factor and corrosion factor which cause failure of pipeline are respectively given. Corrected Bayesian network model is more able to meet the actual conditions, which allows for better practical significance in improving the systematicness, predictability and accuracy in failure quantitative analysis of urban gas pipeline, and also reflects the unique advantage and applicability of the Bayesian network in dealing with complex system risk analysis.