It is recognized that LAN traffic contains highly correlated packet streams and is extremely bursty over a wide range of time scales. No formal stochastic models, like Poisson, batch Poisson, and Markov Modulated Poisson processes, can capture well the burstiness and correlation of LAN traffic. A hierarchical model is proposed in this paper to capture and reproduce LAN traffic. A packet (race of simulated LAN NFS™ traffic is fed into a SMDS access path simulator to evaluate the performance in terms of loss and delay measures. By manipulating a set of parameters, the traffic simulator constructed from this model is able to reproduce various traffic patterns. A burstiness analysis shows that the model can capture well the degree of burstiness over different time scales. The ability of DS1 and DS3 in supporting NFS traffic and the relationship between NFS read/write patterns and performance are also studied using this traffic pattern simulator. An "interval performance criterion" is described as an alternative for performance specification.