A mathematical model is introduced for performance evaluation of an OBS core node employing either no or full wavelength conversion strategies. Furthermore, the model assumes long-range dependent (LRD) traffic arrivals to the OBS intermediate node which are accurately modeled by Pareto distribution. In our proposed model, each output port is imitated by a GI/M/w/w queue for which a single performance measure; namely the burst loss probability, is evaluated from the model equations. Also, results of this model are compared with those obtained while assuming short-range dependent (SRD) Poisson arrivals to the core node in the two cases of no and full wavelength conversion. Finally, results show that traditional Poisson traffic models yields over-optimistic performance measures in terms of lower burst loss probability when compared to the more accurate long-range dependent Pareto traffic model. This discrepancy between the two models is much clearer for light traffic scenarios due to the more significant impact of self-similarity.