Abstract: Based on the meteorological and hydrological observation data, the relationship between annual runoff and meteorological factors in Baihe River Basin is analyzed, and a fitting model is established. The variation characteristics of annual runoff in Baihe River Basin in the 21st century are estimated by using the prediction results of CMIP5 model under different representative concentration pathway (RCPs). The results show that the annual runoff of Baihe River Basin decreased first and then increased from 1981 to 2012, which was negatively correlated with the annual average maximum temperature observed by meteorological stations, and positively correlated with the annual precipitation. A multiple linear regression equation was established with the annual mean maximum temperature and annual precipitation of Hongyuan and Aba stations as independent variables and the annual runoff of Tangke station as dependent variables. The F value test passed the significance test (α = 0.01), and the percentage of the same phase reached 100% in the verification period, indicating that the regression equation was stable and reliable. The prediction of the annual runoff of Baihe River Basin in the 21st century shows that the annual runoff of Baihe River Basin had no obvious change trend under RCP2.6 and could be maintained at the current level. Under the RCP4.5, the annual runoff displayed a decreasing trend of 0.28 × 10⁸m³/10a, which decreased by 17% and 19% in the middle and end of the 21st century, respectively. Under the RCP8.5, the annual runoff displayed a significant decreasing trend of 1.1 × 10⁸m³/10a, and it decreased by 20%, 24% and 49% in the early, middle and late 21st century, respectively.