Abstract: Network arrivals are of ten modeled as Poisson process for analytic simplicity, but recent traffic studies have shown that data-packet interarrivals are not exponentially distributed,but a heavy-tailed distribution, including Pareto distribution. The distributlon of packet interarrivals of the TELNET originator is analyzed, according to the empirical distribution of the Tcplib, and it is shown that packet interarrivals of the TELNET originator can be well modeled by Pareto distribution in large time scales- The characteristics of the burst series and the relationship between the Pareto model and self-similar model and M/G/∞ model are analyzed. It is indicated that the self-similar model is much better than the Poisson model when trying to model the network traffic. This new model can be widely applied in network performance analysis and network performance simulation.