Boehm, Brown, and Lipow 1 have characterized the multidimensional nature of software quality in terms of a hierarchy of attributes. One of the high-level attributes is reliability, which they define qualitatively as the satisfactory performance of intended functions. This definition may be refined to the quantitative statement "probability of failure- free operation in a specified environment for a specified time." A "failure" is an unacceptable departure of program operation from program requirements, where, as in the case of hardware, "unacceptable" must ultimately be defined by the user. The term "fault" will be used to indicate the program defect that causes the failure.