Three models for predicting the gigacycle fatigue strength of high-strength steels were investigated in this study: the Tanaka-Akiniwa model, the Murakami equation and a newly-formulated model. The Tanaka-Akiniwa model proved to have the drawback of overestimating the inclusion size effects. In the Murakami equation, the calculated ODA growth curve was imaginary. Our results demonstrated that the new model generated the most accurate predictions. Predictions were then derived, with each constant calculated by fitting the experimental data to each steel grade. The derived predictions were therefore not universal but different for each steel grade. The accuracy of the derived predictions varied according to the grade of steel. The worst accuracy was for JIS-S40C, the second worst was for JIS-SUJ2 and the third worst was for JIS-SCM440 under R = –1. Other than these three predictions, though, the accuracy was excellent. During the fitting of the experimental data, the existence of new fatigue limits was suggested in the gigacycle regions: these were revealed by fatigue testing up to 1011 cycles. The derived predictions indicated only minor differences between steel grades for gigacycle fatigue strength. With increasing inclusion size, the drop in gigacycle fatigue strength became less steep, and looked at in terms of the mean stress effect, a modified Goodman line could safely be used for high-strength steel.