Aimed at the feedback and dynamic problem encountered in Bayesian network application, we exploit temporal extension to Bayesian network. We presented the definition of temporal Bayesian network, exploited the existence rationality of directed cycle in temporal Bayesian network, and put forward a single node＇ s belief updating algorithm in acyclic temporal Bayesian network based on model simplifying, and a probability inference algorithm to general Bayesian network. Illustration examples were given to show algorithms＇ application