Predicting gear surface fatigue life is vital to gear and transmission design. Although simplified approaches based on the smooth-surface Hertzian contact stress calculations are available, the trend of compact design of high-power gear systems requires the gear design calculation to consider severe operation and lubrication conditions and detailed surface topography. It is needed, therefore, to upgrade the life prediction methods. The research presented in this paper aims to develop a gear pitting life prediction approach based on the rough-surface mixed lubrication model developed by Hu and Zhu  and the fatigue life model developed by Zaretsky and modified by Epstein et al  with accurate surface-subsurface stress analyses. With this approach, gear design parameters, operating conditions, materials, lubricant, and real tooth surface topography are used as input data. When conducting surface fatigue life prediction, three modules are employed: the mixed lubrication module, the stress analysis module, and the fatigue life module. In the mixed lubrication module, machined gear surfaces are digitized and used for calculating the pressure distribution and lubricant film thickness in the mixed lubrication. The mixed-lubrication pressure distribution is inputted into the stress module to obtain surface and subsurface von Mises stresses. These stresses are then integrated in the fatigue life module for estimating the fatigue life corresponding to a certain failure probability (50% in the present study). In other words, the life prediction is now based on subsurface von Mises stresses in the lubricated rough surface contacts, instead of smooth contact Hertzian pressure. There are a few materials constants involved in the fatigue life module. They are calibrated by comparing the prediction results with available Eaton experimental data. Table 1 lists the experimental data for 15 gear sets (most of them are hobbed and shaved) under combined rolling and sliding. The Hertzian stress ranges from 0.8554 GPa to 2.9494 GPa. It has been found that most conventional pitting life prediction methods tend to give conservative life estimates. With optimized materials parameters in the present life model, predicted pitting life results well agree with available Eaton gear test data, and the correlation appears to be 96.3%. Figure 1 shows the comparison between experimental and prediction data. Obviously, the model prediction well represents the performance of this group of gear surfaces in contact and mixed lubrication.