Abstract. In this paper we describe a novel concept for reliability analysis of communication architectures in safety-critical systems. This concept has been motivated by applications in the railway control systems domain, where transitions into stable safe state are usually considered as undesired events because they cause a severe deterioration of the service reliability expected by end users. We introduce a domain-specific language for modelling communication architectures, the protocols involved and the fault hypotheses about anticipated deviations of communication channels and possibly other components from expected behaviour. From such model, a generator creates mutant models associated with probability formulae expressing each mutant’s probability of occurrence. Each mutant is analysed with respect to its unreliability, that is, whether it contains paths leading into stable safe state. Then the system reliability can be conservatively estimated by calculating an upper bound o...