Statistical Methods for Software Reliability Assessment, Past, Present and Future

The role of statistics in software reliability assessment is reviewed in the light of current experience. It is argued that there is a place for statistical methods provided they are founded on the proper sources of uncertainty. These sources are defined and two of them emerge as the basis for reliability predictions. Various perceptions of reliability are identified and attention focuses on one of them, namely the in-use reliability. A general discussion of past and current models for reliability prediction is given together with remarks on several other applications of statistical methods. Some areas of future work are briefly described.