This paper gives a new active queue management algorithm for differentiated service (DiffServ) network, which is based on an new adaptive random early detection (RED) algorithm called pre-estimation RED, so we name it DiffServ PERED. PERED utilizes a short term discrete-time Markov queuing model to estimate the future queue state, then self-adjusts the parameters based on the estimation to gain better performance. DiffServ PERED self-adjusts the parameters not only according to the estimation, but also the priorities of service subscribers, which makes sure that the higher priorities services gain lower drop priorities by utilizing different parameters adjusting functions. In the DiffServ PERED, we present a new idea to control the congestion in network: previous estimation, and a new method to discriminate the services: utilizing different functions to adjust maximum drop probabilities of different priorities services. We demonstrate the performance of Diffserv PERED by comparison between two different priorities services