To solve the problems in the traditional emergency assessment of unconventional emergencies, an unconventional emergency assessment method based on Bayesian network was proposed. Besides, the system efficiency was further improved by the topological optimization of Bayesian network model. Moreover, an example application to a simplified backup power transmission system of nuclear power units was presented to illustrate and test the proposed method. The results show that the proposed Bayesian network based method has valuable characteristics, such as  multiinformation expression,  fusion, and updating,  and can improve the computation efficiency, which make it wellsuited for unconventional emergency assessment.