Abstract The fatigue limits of four rail steels were evaluated with multiple specimens under bending. Inclusion volume and average size were quantitatively measured using an image analyzer in an attempt to correlate these parameters with the fatigue limit. However, no direct relationship was found between the fatigue limits and the elongated inclusion content and average size. Hence, the stress intensity factor at the elongated inclusion tip is considered as a parameter which governs fatigue crack initiation. A theoretical model for the estimation of fatigue limits of the rail steels was developed from the consistency in principle between the definition of the fatigue limit and the fatigue crack propagation threshold. In this model, the stress intensity factor at the inclusion tip was evaluated by taking into account a traction stress at the interface between the inclusion and the matrix. Furthermore, a correction for closure effects was made to obtain the true threshold. The proposed model takes into account the effect of inclusion size and the elastic modulus relative to the matrix, as well as the matrix resistance to crack propagation. As a consequence, a comparison was made between the theoretical values and the experimental values of the fatigue limits.