Currently more and more people are suffering from ever-increasing unexpected e-mails (named spam), as well as the virus and worms introduced by spam. A lot of researchers and studies have focused on spam filtering. In this paper, we firstly present a new anti-spam technique, which include a users' active feedback mechanism, and Maximum entropy based spam filtering approach. Secondly we propose a prototype based on the methods and evaluate the technique by using both well-known mail corpus and real dataset collected from the mail server of our institute. The results demonstrate that the new technique can effectively identify more spam than traditional techniques.