Abstract: A fault diagnosis system of hydropower units is developed using a Bayesian network to consider complicated system structure and uncertain factors in fault diagnosis. This system adopts a Noisy Or model to overcome the shortcomings of the network in large numbers of probabilities and difficulty in determining their values. Taking the characteristic nodes as binary nodes, it uses experts’ knowledge determining the probability at each node and calculates the influence of random combinations of multiple nodes on the outcome, thus determining the possibility of occurrence of a certain fault. The simulations indicate that the number of conditional probabilities required by this new model is decreased from 2n to 2n, and thus it significantly reduces the data demand and raises the time efficiency of hydropower units fault diagnosis.