Abstract: Traditional models of content based information filtering are clumsy to describe the complex relationships of events that affect filtering process. Furthermore, it is difficult for users to interact with the filtering system. To address the above problems, BMIF-an information filtering model founded on Bayesian network is proposed in this paper. It firstly outlines the relationship of features, interests, queries and other main elements in information filtering with a simplified Bayesian network, and then provides six elementary nodes to characterize the relationships in specific conditions. Some use cases of BMIF are presented as well, which includes: ① Describing traditional models with BMIF; ② Adding lexical knowledge to BMIF; ③ Mixing learning with interaction, and combining content based filtering with collaborative filtering.