The cross structures in the South-to-North Water Diversion Middle Route Project face tremendous flood risk. Using Bayesian Network theory and complete learning method, the nodes and structure of the Bayesian Network are first decided and the parameters of each node are determined secondly. The conditional probability table of the cross structure node is determined using the conversion table developed by U.S. Bureau of Reclamation. A Bayesian Network model assessing the cross structure's flood risk is set up finally. This model is applied to evaluate the flood risk of the cross structures above ten typical rivers, provided that the historical "63.8" and "75.8" extraordinary storm which happened near the Middle Route Project occur again. The model's applications in the Project's future flood risk assessment decision support system are discussed at last. The results indicate that the model, which is very valuable, can be applied to evaluate the flood risk of a single cross structure and has a good potential application in the decision support system after completing the Project.