An analytic research on establishing different traffic models on the traffic nature of different time granularity can provide necessary academic foundation for network design and simulation as well as ensuring the quality of service and network management. This paper aims to make simulant predication by means of corresponding math tools on the modeling of real traffic of the different time granularity. The experimental results indicate that the predicated traffic is close to the real traffic distribution. Keywords: Network traffic; Time granularity; Traffic model With the development of Internet, the demands for various network services appear to be higher than ever. Therefore, the exact and sound description of and research on the key nature of network traffic have much significance in network performance management, crowded management and design of the network protocols of good performance. Through the research, it is found that network traffic under different time granularity follows the different behavior rules[1]: the traffic behavior of super-small time at millisecond-level is mainly subjected to network protocol and mechanism, with no displaying of self-similarity; the traffic behavior of wide time granularity at hour-level is a complicated non-linear process and mainly subjected to outside factors, with no self-similarity as well; only the traffic behavior of small time granularity at second-level displays self-similarity. This paper has made a study on network traffic analysis models based on time granularity and made an analysis and prediction about the real traffic.