We consider a stable tandem network which consists of two M/M/1 nodes and study the probability that the total backlog exceeds a large level N. Since the excessive backlog is a rare event, it is difficult to estimate this probability efficiently by using the crude Monte Carlo simulation. Instead we perform the ;$h$-transform proposed by McDonald(1999) to obtain the twisted network, in which the node with the larger load is overloaded. Then we use it to run the fast simulation.