Abstract: The stress-energy coupled criterion based on the finite fracture mechanics is widely used to evaluate the notch fatigue limit, as it overcomes the disadvantage that the classic linear elastic fracture mechanics is not suitable for solving the crack initiation of notches. However, the prediction results of this method are often larger than the experimental results. Therefore, it is necessary to modify the coupled criterion. It is clear that the cracks initiated at notches are actually fatigue short cracks. And the stress-energy coupling method adopts the long crack threshold to calculate the equivalent stress intensity factor, which is inconsistent with the actual situation. Hence, the material parameter of the short crack threshold is incorporated into the energy equation to modify the coupled criterion. In order to determine the value of the notch short crack threshold, the Meggiolaro notch short crack threshold model was employed, and the optimal value of the parameter in the Meggiolaro model was determined based on a large number of experimental results. To verify the effectiveness of the modified method, the fatigue limits of 50 groups of notched specimens were predicted and compared with the experimental results. The results show that the average prediction error is significantly reduced from 9.9% to 3.9% after modification, indicating a substantial improvement in prediction accuracy. Meanwhile, the proportion of positive prediction errors (i.e., predicted values are higher than experimental values) decreases from 78% to 60%, making the prediction results more neutral and effectively improving the shortcoming of non-conservative predictions in the original method. Further analysis results reveal that as the notch radius decreases, the stress intensity factor calculated based on the notch fatigue limit and the initiating crack becomes closer to the short crack threshold, leading to smaller prediction errors.