Experimental determination of strain-fatigue life curve is expensive considering the time and effort involved. Several empirical relations based on monotonic tensile properties and/or hardness is available for estimating strain-fatigue life properties. Existing criteria to evaluate the accuracy of these empirical relations are limited by the data points in the strain-fatigue life curve. A new criterion based on the prediction errors of fatigue constants is proposed to overcome the limitation. Around eight prediction methods are evaluated for two hundred twenty-six steel grades from literature using the existing and the proposed criteria. It is observed that the conclusions from the new criterion are more reliable than the existing criteria. © 2010 Elsevier Ltd. All rights reserved.