The burst loss probability is considered as one of the most important performance indicators of optical burst switching (OBS) networks. Computer simulations are widely used in estimating burst loss probabilities in OBS networks, especially when analytical methods are not possible. However, as the number of wavelengths in an OBS network becomes large, burst losses can become rare events. In such cases, traditional simulation methods may take an exceedingly long time while still not yielding an estimate with acceptable confidence interval. In this paper, we propose a method based on Importance Sampling (IS) to accelerate the simulation. To optimize IS parameters in such an environment, we propose using Simulated Annealing (SA) to directly minimize the variance of IS estimators. The proposed method (SA-ISSC) is easy to use and efficient, while producing very favorable results.