Abstract: The earliest observed meteorological data in western China archived in China Meteorological Administration began in the 1930s, which cannot meet the users' needs for the establishment of century-scale climate change series over China. Meanwhile, the reconstruction of paleoclimate proxies or climate model simulations can be extended to the time before the instrumental era. In order to investigate the merging approach of long-term, multi-source climate data, the Bayesian approach was used to merge the tree-ring reconstruction, observations and the CMIP5 model surface air temperature data. The temperature proxy data were calibrated and reconstructed first based on observations, and were used as a prior distribution of the Bayesian model. Several models that have the best modeling effect were then selected based on Taylor diagrams. Finally, the Bayesian model was used to merge the paleoclimate reconstruction dataset and climate modeling series. The results show that the Bayesian model can effectively extract useful information from various data sources for merging. The long-term change (linear) trend of the merged series tends to be closer to the observed climate series, and the accuracy of the series was improved to a certain extent, and the uncertainty of the results was reduced. Moreover, the merged results can in a certain degree correct obvious deviations of the prior distribution and the climate model data, respectively. Results of the present study provide a feasible idea for the reconstruction of long climate time series.