The exact evaluation of usual reliability measures of communication networks is seriously limited because of the excessive computational time usually needed to obtain them. In the general case, the computation of almost all the interesting reliability metrics are NP-hard problems. An alternative approach is to estimate them by means of a Monte Carlo simulation. This allows to deal with larger models than those that can be evaluated exactly. In this paper, we propose an algorithm much more performant in time and in precision that the standard Monte Carlo technique. Moreover, it is particularly efficient in the case of highly reliable systems. It will be shown that it behaves much better than the so called dagger sampling plan on which good results have been reported in the literature. We will also show that the applicability of the dagger method depends on the reliabilities of the components of the network while this is not the case of the method proposed here.