The cyclic strain energy method has been widely used because it can predict fatigue life from the perspective of fatigue damage mechanism and explain many fatigue phenomena which cannot be explained by other methods. However, in the large amount of empirical study of the overall properties of materials, the machined surface layer is often not considered, which affects the prediction accuracy of the energy method. Based on the energy method, a new fatigue life prediction model with back stress energy based on the machined surface layer of high strength steel was proposed. The effect of surface mechanical characteristics on total back stress energy was considered by introducing the area of residual stress with depth as the influence factor, while the effect of surface layer geometry and metallurgy on strain energy was considered by taking the non-propagation threshold of microcrack as the influence factor. The results showed that: the dispersion band of fatigue life prediction error for the modified model considering surface integrity was reduced by 38% and the prediction accuracy improved by 25%; it had the same error dispersion band (1.25 times) as the model for the real-time statistical single cycle energy density after fatigue fracture, which overcomes the disadvantage that the fatigue life can be predicted only after the fatigue test of the traditional energy method; for quenched and tempered high strength steel, the average accuracy of fatigue life prediction of the modified model considering surface integrity was improved from 72.7% to 90.6%; compared with the real-time statistics after fatigue fracture, the error dispersion band of the single cycle energy density prediction model was reduced from 3.30 times to 1.41 times. This model provides a method to assess the fatigue service performance of the machined surface layer, and improves the applicability of strain energy in different machined surface layer characteristics.