Abstract: The GRAPES (Global-Regional Assimilation and PrEdiction System)_Meso model developed by China Meteorological Administration is coupled with a hydrological model to increase lead-time of flood forecast. GRAPES_Meso model is run in 15 km×15 km horizontal resolution and 5 km×5 km horizontal resolution. The initial fields and lateral boundaries of 15 km×15 km horizontal resolution of GRAPES is provided by global NCEP forecast datasets, and the initial fields and lateral boundaries of 5 km×5 km horizontal resolution of GRAPES is provided by 15 km×15 km horizontal resolution of GRAPES. In order to match the input scale of hydrological model, quantitative precipitation forecasts of GRAPES_Meso model is downscaled to 5 km×5 km horizontal resolution. Xin'anjiang model and grid-based distributed Xin'anjiang model are used, which have been widely applied and proven effective in flood forecasting and hydrological simulation in humid and semi-humid regions of China for a long term. Wangjiaba Station and Xixian Basin in the upper reaches of the Huai River are chosen as sensitive areas. The two hydrological models are driven by forecast datasets of GRAPES. Upstream Wangjiaba Station, the basin is divided into 10 sub-basins for the coupling experiment of Xin'anjiang model. And Xixian Basin is for the coupling experiment of grid-based distributed Xin'anjiang model. A flood which maintains from 0800 BT 28 August to 1400 BT 29 September in 2009 is forecasted by these two models. The experiment results show that compared with observed precipitation, quantitative products of GRAPES model in 15 km×15 km and 5 km×5 km horizontal resolutions are well consistent. The quantitative products of GRAPES model with 5 km×5 km are larger than the quantitative products of GRAPES model with 15 km×15 km. A promising tool is given by GRAPES meteorological and hydrological coupled hydrologic model to increase lead-time of real-time flood forecast, compared with that driven by raingauge observation. The accuracy of the flood forecasting based on the precipitation prediction of GRAPES model is approximate to the precipitation prediction. The performance may be better if the input requirements for hydrological models are exactly met.