The new generation mesoscale model WRF is used to simulate heavy rainfall in the Changjiang-Huaihe valley during 7－8 July 2007, and the precipitation sensitivity to land surface parameterization schemes is tested. All sensitivity simulations are initialized with NCEP-FNL data (1°×1°) . The results show that the simulated precipitation patterns from the WRF model generally agree with the observations, and the sensitivity of the precipitation simulation to the land surface physical process is distinct. Coupling the surface process parameterization scheme can improve the precipitation simulation significantly. However, the simulated precipitation with different schemes is lower than the observed precipitation, and details of simulated precipitation patterns, amount, diurnal variation and precipitation types are different among the land surface parameterization schemes. Comparing all results of precipitation simulation with different land surface parameterization schemes, NOAH scheme gives more reasonable results than the others, and when the simulated domain is urban, UCM scheme is superior to the others. Sensitivity experiments indicate that understanding and appropriately using land surface parameterization schemes are necessary to enhance the modeling capability.