Since the fatigue phenomenon is characterized as a probability feature and test data cannot be easily collected number its small sample data are uncertain. Therefore, statistical evaluation methods should necessarily be introduced into data evaluation. With such a basic concept in mind, this study attempted to apply conventional statistical processing methods to the fatigue data and thereby present a new evaluation method in consideration of the fact that the fatigue test is usually performed within a limited number. That is a package evaluation method was adopted which co-realtes parameters between different levels of capacity in E-N or S-N level. So far fatigue limits have been found by means of stiarcase method. But it was also disclosed that this new method has its own disadvantage because limited frequencies are not duely considered.