Computer networks are becoming an important aspect of our life. More users are competing for limited resources; degradation in the quality of service is thus unavoidable. Proper modeling of bursts and serious deterioration is important. We model network traffic bursts using generalized Pareto distribution. This method is based on the extreme value theory. We simulated external and collected internal network traffic traces. We found that GPD models these two types of traffic correctly and with clear differences between them. New measures and metrics can be defined out of the GPD modeling to be included in the service level agreement between provider and users. This will have the impact of enhancing in the expected quality of service, especially for mission critical applications. © 2010 IEEE.