Future high speed networks using the Asynchronous Transfer Mode will be able to carry a broad range of tra c classes and will also have to provide a quality of service (QoS) measure to many of the tra c classes. Typical QoS measures are cell loss probability (CLP) and cell transfer delay (CTD), which includes queueing delays and service times.For ATM networks with heterogeneous input tra c, namely constant bit rate (CBR) and variable bit rate (VBR) tra c, we consider the aggregate probability that the cell transfer delay exceeds a given threshold (delay threshold probability). We use the connection tra c descriptors standardized by the ATM Forum to characterize the input tra c. To estimate the rare delay threshold probability, we rst present a multinomial Monte Carlo simulation model. The multinomial formulation e ectively removes the correlations associated with the bursty events. We then present a three part Importance Sampling procedure to increase the e ciency of the simulation. For the experimental systems considered, the improvement in simulation e ciency is inversely proportional to the probability being estimated 1 .