The objectives are to investigate the fundamental reasons which cause independently developed software programs to fail dependently, and to examine fault tolerant software structures which maximize reliability gain in the presence of such dependent failure behavior. The authors used 20 redundant programs from a software reliability experiment to analyze the software errors causing coincident failures, to compare the reliability of N-version and recovery block structures composed of these programs, and to examine the impact of diversity on software reliability using subpopulations of these programs. The results indicate that both conceptually related and unrelated errors can cause coincident failures and that recovery block structures offer more reliability gain than N-version structures if acceptance checks that fail independently from the software components are available. The authors present a theory of general program checkers that have potential application for acceptance tests. Walker, J. Leslie and Caglayan, Alper K. Unspecified Center...