Abstract: The rock mass rating system (RMR) plays a vital role in geomechanical evaluation.The traditional rock mass quality classification uses a combination of experiences and geostatistics, and the results are often not accurate enough, especially in areas with weak strata such as faults and caves.In order to solve the inherent uncertainties and errors of these empirical techniques, this paper proposes a progressive RMR prediction strategy based on Bayesian framework.The strategy includes three key components: ① Variogram modeling.We use the observation data of the excavated surface to construct and update the variogram model to capture the spatial variability of RMR.② TSP-RMR statistical model.The TSP-RMR statistical model is integrated into the Bayesian sequence update process.③ Bayesian maximum entropy (BME) integration.The geological information obtained by excavation is combined with tunnel seismic prediction (TSP) data by BME method.The proposed method was applied to a tunnel engineering practice.The results show that the fusion of TSP data and geological exposure data significantly improves the accuracy of RMR prediction.On the selected prediction points, the relative error of this method is less than 15% compared with the traditional Kriging method.