UltraSAN based on stochastic activity networks (SANs), provides a sophisticated model-based environment for evaluating large-scale telecommunications systems. When modeling telecommunications systems, one must often contend with performance measures based on rare events. Such performance measures pose a difficult problem for traditional model solution techniques. In particular, performance measures based on rare events lead to stiff systems that are difficult to solve analytically. Traditional simulation techniques can take a very long time to collect a statistically significant sample of a performance variable based of a rare event. The efficiency of simulation-based evaluations of such measures can be improved by using a variance reduction technique called importance sampling. The authors discuss the UltraSAN software and a general framework developed for the application of importance sampling to SANs within the UltraSAN environment.