The probability of failure is treated as a statistical quantity from the viewpoint that the probability can be evaluated only through experimental data. A distribution function for the probability of fatigue failure is derived under the linear damage accumulation hypothesis, and it is pointed out that the allowable probability of failure varies largely with its reliability.
A simple but important policy for fatigue-proof design is proposed. This policy requires that a value smaller than a pre-specified allowable value for the probability of failure be correct with a pre-specified reliability.