A true application-specific Network-on-Chip (NoC) requires the exact knowledge of the internal traffic behavior during the NoC is fully running the target application. Therefore, an accurate traffic measuring method is necessary. In this work, a simple but powerful traffic monitoring system is presented for the accurate evaluation and refinement of an application-specific NoC. The traffic monitoring system measures various traffic parameters such as an end-to-end latency, a queuing buffer usage and a congestion level. A NoC-based portable multimedia system as a target system is implemented on multiple FPGAs to demonstrate the effectiveness of the proposed traffic monitoring system. Using the traffic monitoring system, the target system is diagnosed and refined in two different topologies; mesh and star topologies. For the application-specific NoC, three design refinement processes such as buffer size assignment, network frequency selection, and run-time routing path modification are performed in the target system. As a result of the refinement, 42% buffer reduction and 28% latency reduction are obtained in the mesh-connected application-specific NoC.