With the development of the Internet and network services, broadband user behavior tends to diversify. It is a rising challenge for the network operators to understand network user behavior. In this paper, we use the Postgres to sort the data records which are caught and extracted from the network monitoring equipment. Then we analyze the characteristics of the data by SPSS, such as login time, logout time, on-line duration and log times and the relationships among them. We extract the main feature, and then use the K-MEAN algorithm to classify the users and carry on the corresponding description. Our results provide a basis for network operators to understand dial-up user behavior, optimize network planning and adjust customer.