Abstract: Soft real-time applications may be rate-adaptive or delay-adaptive, showing much flexibility in bandwidth requirements. For such traffic it is acceptable to allow users instantaneous rate to be more or less than what heshe requires. Taking this utility characteristics into account, a network pricing mechanism with congestion charge and compensation is proposed, which differs from the traditional congestion pricing in that users traffic is distinguished so as to punish the misbehaving users by extra congestion cost and compensate the innocent users by price discount while network congestion is onset. Following this idea, the rules for setting congestion charge and compensation are examined in detail. By comparing the acquired transmission rate T\+* with the pre-agreed rate T\+0 negotiated during admission control and the actual arrival rate T, the network users are classified into three kinds so that the basic connection price, congestion charge and compensation could be applied respectively. Then a dynamic pricing algorithm is presented accordingly along with the analysis on the incentive compatibility of charging policy. Simulation results show that the pricing algorithm with congestion charge and compensation is reasonable in that users are charged appropriately in accordance with the pre-agreed service level and actually allocated network resources. With this pricing mechanism, users would like to announce their true requirements through service negotiation while network nodes would like to participate in the pricing game as well to make more profit.