Abstract: The recent measurements have shown the existence of long range dependence and self similarity in real network traffic. Some researchers use FGN （fractional Gaussian noise） as a traffic model for this sort of traffic, but FGN can only capture the long range dependence. In this paper, a new method is suggested, which uses the so called FARIMA （fractional autoregressive integrated moving average） to modeling network traffic. The method of building the model is given in implementation detail. FARIMA （ p,d,q ） model is a good traffic model that is capable of capturing both the long range and short range behavior of a network traffic. The method is also applied to the real network traffic data. The experiments show that FARIMA model could be used to model actual traffic on quite a large time scale. Compared with other short range dependent processes such as ARMA models, less parameters are required by the FARIMA models. The performance comparisons between FARIMA and other traditional models such as AR, ARIMA and FGN are presented, and the results show that FARIMA model is better than other traditional traffic models. Future work includes applying FARIMA model to various long term and/or short term traffic predictions.