The reliability at time t of a system in sustained operation is often taken to be the probability that it functions continuously during the time interval (O, t). The standard computation of system reliability finds the probability that the system functions at time t in terms of the probabilities that its components function at time t. This procedure is relevant only if the system, and its components, have lives (roughly speaking, a device has a life it it functions continuously until some time of failure, and remains failed thereafter). It is shown that if each component of a coherent system has a life, then the system has a life (again roughly, a system is coherent if its performance is not impaired by an improvement in the performance of its components). The principal result is that, under reasonable conditions, the converse is true: if the system has a life, then the system is coherent and each component has a life. This means that if the standard computation of system reliability is to be used, the system in question should be coherent. (Author).