A typical operating environment of a packet switching (store-and-forward) computer communication network is that it is shared by many users with different classes of packets. Packets may be classified in a very general fashion by types of users, messages, applications, transactions, response time requirements, packet parameters such as packet rate and length, and by network parameters such as source-destination and path length. A well-designed network must provide access and performance assurance to all packet classes. An architecture for designing such a communication network is presented in [1]. This architecture is based on MPLS as an underlying technology used for IP transport and DiffServ-like mechanisms for QoS provision which use simple priority queuing to differentiate several traffic classes by assigning them different delay targets. In this paper, we introduce a mathematical model for performance evaluation for QoS Provisioning in MPLS/DiffServ-Based IP Networks. This mathematical model is formulated to evaluate the performance of the proposed architecture described in [1]. Two parts are considered in the evaluation. One, is the analysis of the proposed QoS routing algorithms that are used in the proposed architecture, the other is the analysis of the average packet delay over the network. An M/G/1 Non-preemptive priority queuing system is used as the evaluation model to analyze the average packet delay over the network. Relative MPLS/DiffServ with Non-Preemptive Priority Queuing system performance compared to MPLS/DiffServ without priority (FIFO)[2] is evaluated. The results indicate that for well-designed MPLS/DiffServ equipments, original IP packet delay can be reduced by 39% or more with FIFO queuing system [2] and 42% or more with Nonpreemptive priority queuing system