This study focuses on the method of predicting the fatigue life of materials subjected to random loading. Since random stress caused by random loading is rigorously expressed as stress power spectral density (PSD), the fatigue life should be predicted with stress PSD in frequency domain. We therefore propose a method of improving the accuracy of the fatigue life prediction in frequency domain. Although the method proposed by Dirlik is widely used for predicting the fatigue life in frequency domain, it often overestimates fatigue damage. In particular, the fatigue damage caused by large stress amplitude is overestimated, and hence we set the upper limit of stress amplitude to be taken into account. The proposed method was applied to illustrative examples, and as a result, it worked well to improve the accuracy of the fatigue life prediction. By using presented method for predicting fatigue life in frequency domain, we can develop lightweight transportation machinery, such as trains, subjected to random loading.