Abstract: As the source congestion in TCP network plays an important role in causing the network instability and deteriorating the quality of service, this study aims to investigate a robust congestion control to solve this problem. A fluid-flow model was formed to represent the nonlinear TCP behavior and the Sliding-Mode Prediction (SMP) approach was applied to design the active queue management controller. The stability and robustness of the SMP scheme are then effective validated under different network scenarios in NS-2. Simulations results verify that SMP scheme can provide smaller queue fluctuations, higher link utilization and low packet loss rate than the traditional controllers.