Development of fault-tolerant computing systems requires accurate reliability assessment techniques. Usually, the reliability measures are functions of component failure rates and fault coverage probabilities. Coverage provides information about the fault and error detection, isolation and system recovery capabilities. This parameter can be estimated by physical or simulated fault injection. One of the most difficult problems the analyst has to deal with, throughout the fault injection process, is the largeness of the fault space. This paper addresses the problem of inferring the coverage probabilities from the information gathered in physical or simulated fault injection experiments. A 3-stage sampling technique is developed for coping with the largeness of the fault space. Statistical experiments are carried out in a three-dimensional fault space which takes into account the inputs applied to the system, fault occurrence times and fault locations.