This paper explores and evaluates several aspects of quality of service (QOS) in an integrated traffic ATM network. Specifically, we consider the relationships between service class definitions, usage parameter control (UPC) or policing of customer traffic, network resource allocation mechanisms, and specific network conditions under which realistic QOS limits on cell loss ratio (CLR) can be met. Traffic consists of a mixture of voice, video, image and data divided into two service classes for UPC and network resource allocation. The block oriented network simulator (BONeS) tool is used for performance evaluations. Limitations associated with evaluating very low CLR values (e.g. and via simulation are overcome by a hybrid simulation and extrapolation technique. It is found that these stringent QOS limits on CLR can be met for most traffic with the techniques used here, reasonable backbone trunk loads (75 per cent) and reasonable buffer sizes (200-250 cell buffers per queue). However, meeting similar limits for extremely bursty traffic (such as our image traffic model) would require extra care in network design and operation, including judicious segregation of traffic.