Abstract: Though there exists many defects in traditional cognitive reliability and error analysis methods (CREAM) , which leads to narrow application, they still have great development potential. This paper proposes a model combining an improved weighted Bayesian network (BN) and CREAM to improve the accuracy of evaluation and expand the application of the method. In this model, the weights of factors in the scenario are obtained first by the Grey-DEMATEL method. Then, the probability distribution of adjusted common performance condition (CPC) factor nodes is obtained by Bayesian network inference. Afterwards the weighted factors are sampled and simulated to obtain the probability distribution of the control mode in the scenario. A case study of 1800 W customized SEED power supply is used to verify the effectiveness and feasibility of the model through analysis of evaluation results and comparison with other methods. The improved model has wider applicability and more objective outputs, which provides a reference for the construction of Human Reliability Analysis (HRA) index systems.