In MPLS networks, CR-LDP (Constraint-Based Routed Label Distribution Protocol) provides traffic engineering and QoS (Quality of Service) by distributing labels along the path. One of the innovative ideas in CR-LDP is right in the capability of bandwidth preemption. Bandwidth preemption allows a high-priority traffic flow to pre-empt the low-priority traffic flow when there is not enough bandwidth for conveying the high-priority flow. However, it is not effective in assuring QoS by simply pre-empting the low-priority flow. Therefore, in the thesis, we present a bandwidth preemption with negotiation (BPN) architecture for MPLS networks. In BPN, a high-priority flow will have to negotiate the bandwidth with Ingress Switch Router (ISR) before it can actually pre-empt the bandwidth being used by the low-priority flow. A network link-state database in ISR is designed to record the remaining bandwidth for each priority class. ISR determines whether a high-priority flow can pre-empt a low-priority flow based on the condition that the low-priority flow is possible to switch to other paths with an equal bandwidth along the path. In order to evaluate the performance of our proposed BPN mechanisms, we modify the MNS (MPLS Network Simulator) by adding a bandwidth negotiation module. For the comparisons, we design two topologies for simulating the proposed BPN and the traditional bandwidth preemption with force mode. It is observed that the BPN have exhibited better performance in average throughput and packet loss rate than the traditional bandwidth preemption, not matter either a complex or a regular topology is used. Finally, we analyze the BPN algorithm complexity by some network parameters, and compare the complexity with that of traditional bandwidth preemption