It has very realistic significance for improving the quality of users' accessing information to filter and selectively retrieve the large number of information on the Internet. On the basis of analyzing the existing users' interest models and some basic questions of users' interest (representation, derivation and identification of users' interest), a Bayesian network based users' interest model is given. In this model, the users' interest reduction algorithm based on Markov Blanket model is used to reduce the interest noise, and then users' interested and not interested documents are used to train the Bayesian network. Compared to the simple model, this model has the following advantages like small space requirements, simple reasoning method and high recognition rate. The experiment result shows this model can more appropriately reflect the user's interest, and has higher performance and good usability.With the rapid development of Internet and the World Wide Web, information on the Internet has greatly enriched. It is a very significant matter to filter and sort the information according to users' interest [1][2][3][4][5][6] . Therefore, we need to establish an appropriate users' interest model and to build a suitable users' interest mining algorithm.The simple interest model is a simple method describing the users' interest [3][4][5] . In this model, the users' interest is represented as an interest set. The set of all interest makes up of the full set (the dictionary). The full set of interest is denoted as 1 2 { , , , }, m T t t t where 1 2, , , m t t t represent the interest (lemma) respectively, m is the size of the dictionary T. Based on the interest set, some basic concepts are first given for the easy discussion.Definition 1 The interest node is defined as a pair: Node (t) = (t, weight), where t is the lemma in the interest set T and weight is the weight of t.The text set in the WWW cache is denoted as D, D=For the lemma i t in the dictionary, its frequency appearing in the document j d (term frequency) is denoted as tf ij ; and its all frequency (no matter how many times it appears, it is recorded as 1) is denoted as df j (the document frequency). The reciprocal of df j is named as the invert document frequency, and is denoted as idf j .For the simple interest model, it is an important aspect to compute the weights of the interest nodes. The data mining of the simple interest model is mainly the text based data mining [7,8] . In its interest computing method, all the documents in D are looked as one hyper-