A grid-based distributed hydrological model (three different grid sizes : 1, 2, and 4 km) is used to study the impact of grid size on hydrological process modeling in a large catchment of Xinjiang River, China. Results from the study indicate that scaling of grid size absolutely affects the partition of water dist ribution between surface runoff and subsurface flow, with modeled base flow increasing obviously as grid size increases. Modeled annual stream flow shows a slight decrease as grid size increases. Scaling of grid size has no significant effect on actual evapot ranspiration of the catchment, but water infilt ration increases obviously when surface slope becomes flatter with a bigger grid size. After calibration, the three grid size models all have a well performance in modeling stream flow, but the 2 km grid size model performs better than the other two models. From this study, it can be concluded that fine resolution of grid size does not mean to improve the modeling accuracy in a distributed hydrological model and coarse resolution of grid sizes can be used for stream flow modeling, depending on watershed size and the level of accuracy required. For a specific resolution of soil data, there is an optimum grid size at which the model performs best .