Generalized fatigue damage is derived based on S-N curve under constant amplitude loads with S characterizes the load cycle by the combination of load peak and load amplitude. For random fatigue loads, a natural choice is to use half cycle (Range) counting method. Rain Flow counting is popular due to its useful nature to pick-out full cycles among random loads. It is found in the present research through analysis and verified by experimental results that Rain Flow counting seriously overestimate the damage for the actual true random loads owing to its inherent nature and the Range counting method is more reliable. As the filtering threshold increases gradually, the overestimation effect by Rain Flow counting method becomes less, and the calculated damages by Rain Flow and Range counting begin to converge. The filtering threshold damage corresponding to the “fatigue limits” for the concerned case was calculated. According to equal damage principle, the filtering threshold amplitudes were derived and found to vary with the load peaks. In practice, an acceptable solution is to use quasi-variable filtering thresholds.