Abstract: This study implements ensemble runoff forecasts using the Xin'anjiang model driven by ECMWF,UKMO,NCEP and other seven control forecast products of TIGGE data center in Chitan reservoir watershed in Jinxi,Fujian Province,China. By means of ensemble selection,pre-processing via multi-model integration,and post-processing based on the BMA model,the influences of data-processing scheme and initial set quality on the accuracy and uncertainty of the meteo-hydrological runoff predictions were investigated. The results show that different treatment schemes could effectively improve the accuracy and stability of runoff forecasts. Combination of pre-processing and post-processing led to better performances than other schemes due to its error reduction in two aspects,i.e. lowering the input errors and controlling the output errors. Although the initial set quality exerted a perceptible impact on the ensemble runoff forecasts,the effect was not significant since either pre-processing or post-processing procedure effectively controlled the forecasting errors. Overall,we conclude that the pre-processing and post-processing processes are indispensable to improve the accuracy and reliability of meteo-hydrological runoff forecasts,which should be paid attention to.