Abstract:A modified semi-distributed monthly water balance model is proposed to simulate the streamflow in the Tarim River. With the comparative study among TOPMODEL, Xinanjiang model and the modified semi-distributed monthly water balance model in the headwater catchment of the Tarim River, it showed that the proposed model with simple structure and two parameters as well as TOPMODEL, Xinanjiang model perform well in the study area (the Nash-Sutcliffe efficiencies R 2 were between 0Ð60 and 0Ð69, and the relative errors of the volumetric fitness R E were between 19% and 19%). At Shaliguilank hydrological station in the Aksu River basin, R 2 is around 0Ð60. Those values at other stations are generally between 0Ð60 and 0Ð66 in headwater catchment of the Tarim River. The high-flow year, average year and dry year were determined by analysing the frequency of annual runoff in the Tarim River from 1957 to 2005 using Pearson type III (P-III) distribution, and the L-moment approach was applied to the parameter estimation for P-III distribution. With the reference years (high-flow year, average year and dry year) and the long-term trend, different scenarios for the precipitation in 2010 and 2020 have been generated. The streamflow was simulated by using the proposed model with different scenarios, and the impact of climate change on streamflow was analysed. The results showed that the streamflow in