Based on generalizing the results of numerous experimental investigations, the empirical relationships between the fatigue limit of steels and the stress gradient are proposed and justified, which take into account the fatigue damage behavior of smooth specimens and specimens with stress concentrators. A method for determining the parameters of these relationships with consideration of the mechanical properties of steels is proposed. Good agreement of the proposed relationships with the experimental results is shown.Introduction. Numerous experimental investigations on high-cycle fatigue of metals and alloys show a significant increase in the fatigue limit with an increase in the stress gradient that characterizes the nonuniformity of the stress distribution over the specimen cross section.Thus, the fatigue limits in bending are considerably higher than those under axial loading, the local stresses at the stress concentrator tip that correspond to the fatigue limits are higher than the fatigue limits in the uniform stressed state.The generalization of the investigation results is done either by constructing the empirical relationships that relate the fatigue limit value to the stress gradient [1], or by constructing the models that are based on certain hypotheses.It is supposed in [2] that the difference in the fatigue limits in the uniform and nonuniform stressed states is governed by the difference between the nominal stresses calculated on the assumption of elastic deformation and the actual ones calculated taking into account the inelastic cyclic deformation in the nonuniform stressed state.In the studies described in [3, 4, and others], the stress gradient effect is explained in terms of statistical strength theories.In the investigations [5, 6, and others], the explanation of this effect is given in terms of critical distance theories, which assume that either stresses at some distance from the surface or averaged stresses in the surface layer are responsible for the fatigue fracture.In [7-10, and others], the stress gradient effect is due to the difference in the inelastic deformation behavior of surface layers of the metal under conditions of uniform and nonuniform stressed states.All the mechanisms that underlie the above hypotheses take place, to a certain extent, in real materials during fatigue fracture. With the use of the relationships that are based on one of these hypotheses, the consideration of the effect of other factors is achieved by the empirical selection of parameters in the chosen model.Empirical relationships that relate the fatigue limit to the stress gradient are the most simple to use because there is no need to determine the parameters corresponding to different hypotheses, which is not an easy task.At the same time, empirical relationships, many of which have been formulated quite a long time ago, need to be further improved taking into account new experimental data, with new approaches to generalizing experimental results being developed.