A comprehensive classification of failures in computing systems is discussed. The underlying guidelines are: (1) a failure classification must be independent of any particular system; that is, it should be able to be applied to every system; (2) a failure classification must be as detailed as possible; (3) treatment of failures following detection and other parameters, such as the severity of failures, are to be considered as a second step, since they may require a defined categorization. The main advantage of the failure classification is the possibility of characterizing systems with respect to their failure modes, providing designers with a way to choose the most appropriate detection techniques for each particular system. >