Network traffic measurement is one of the most important elements in network management. Without knowing the network status and understanding the behavior of network users, a fast and stable network service cannot be provided. Traditional surveillance based system requires huge data storage to store the traffic information; SNMP and RMON can only provide limited and the predefined metric information. The flow-based traffic measurement system has many advantages over these traditional packet based traffic measurement methods. This paper proposed a Flow-based Traffic Measurement System (FBTMS). The traffic flow is defined based on the packet train model. In this Flow-based Traffic Measurement System, flow is defined as a bi-directional sequence of packets between given source and destination endpoints. The endpoints are identified both by IP address as well as by the port number of transport layer, and flows are divided into TCP based flow and timeout based flow, The packet matching is one of the most important issues in FBTMS, the system must be able to identify whether a packet is belonging to an existing flow efficiently. The system also needs to manage the flow table effectively and remove inactive flows before the flow table getting full. FBTMS includes five components FlowMeter, FlowCollector, FlowConfiguration Manager, FlowServer and FlowAnalyzer. The FBTMS is defined as a hierarchical distributed system, with fully redundant design, active and passive data collection, dynamic and flexible timeout parameter for flows and network traffic analysis in global view.