Accurate prediction of fatigue crack growth serves as the cornerstone for aircraft component lifespan monitoring and residual life estimation. In this paper, a prediction method of structural crack propagation based on dynamic Bayesian network is proposed, which combines the prior knowledge and the posterior knowledge of fatigue crack propagation to accurately infer the crack length. The influence of different particle numbers on the inference accuracy of the dynamic Bayesian network was studied. Through the study of crack propagation of the single hole plate and the lug under random load spectrum, it is shown that the dynamic Bayesian network method can accurately predict the crack growth of structures, and the prediction accuracy is more than 50% higher than that of the traditional method.