To overcome the limitation of the classical stage-discharge relationship model in describing the dynamic characteristics of a river,the locally weighted regression method was used to estimate the model parameters. In order to improve the estimation precision and the calcu -lation efficiency of river discharge, a novel method called clustering-tree weighted regression was proposed. Firstly, the trained samples wereclustered in this method. Secondly, k-nearest neighbors method was used to cluster new stage samples into the best fit clustering. Finally, thedaily discharge of the river was estimated. During the estimation process, the interference of irrelevant information was avoided, so the estima -tion precision and efficiency of daily discharge were improved. The data observed at some hydrological stations were used for the test. The sim -ulation results show that the estimation precision of this method is high. This provides a new effective method for the estimation of parameters ofstage-discharge relationship model.