This paper proposes an AIFS-based multi-class model to study the saturation channel throughput on the basis of the operation mechanism of IEEE 802.11e EDCA (enhanced distributed channel access) supporting servicedifferentiation. The proposed model is calculated numerically and validated against simulation results, a good matchbetween the analytical model and simulation is observed. In particular, this paper compares the proposed modelwith Xiao’s Markov chain model under the same configures. Based on the proposed model, this paper researches thequasi-optimal condition to avoid the computational complexity but still maintain the channel throughput close to itsoptimal value. Finally, the paper proposes an adaptive p-persistent MAC scheme, named DPS (dynamicparameter-tuning scheme), to assign appropriate different transmission probabilities (or contention window size) of different classes. Through DPS, it is feasible to provide service differentiation and achieve targeted throughput ratioamong different classes, at the same time to maximize the total channel throughput. Simulation and numericalresults show that DPS can effectively achieve the performance goal under a variety of network conditions, and achieve higher channel throughput than standard IEEE 802.11e EDCA in all different environments.