The authors discuss how an asymptotically efficient simulation technique, originally proposed for binary communication systems, can be applied also to modulation systems with a large number of signalling waveforms, retaining much of the efficiency of the method. Basically, the probability of error is decomposed into sums of terms corresponding to all possible errors. Then each term is evaluated by a series ofad hocexperiments, designed as if the modulation system were binary. For a given computational effort, each experiment is performed an appropriate number of times to minimise the variance of the estimated probability of error. When the number of waveforms is very large, we show that random selection of the experiments can provide an unbiased form of stochastic truncation of the sum. The paper presents examples of application to block codes, convolutional codes, and trellis coded modulation. These can be easily handled, provided sufficient information about the code structure is available.