There are severe problems in using current software reliability models in safety-critical applications. One of the main reasons is that there are many models with few convincing validations. Validation is particularly weighty with models based on an accurate analysis of the program structure. It would require the application of the model over a large number of software programs and then an accurate and extensive testing of these programs to determine their actual reliability, but the effort required for the program analysis makes the use of a large sample of programs too expensive. This paper shows an approach in which several faulty programs were derived by seeding faults in a correct program. The paper describes the results obtained using this approach to validate a reliability model for critical software.