A number of studies have demonstrated effects of the landscape composition and spatial configuration on the runoff, sediment yield and non-point pollution in a watershed. The SWAT model(Soil and Water Assessment Tool) was widely employed to explore the significance of land management practices on water, sediment, and agricultural chemical yields, which subdivides a watershed into sub-basins connected by a stream network and contain many Hydrologic Response Units (HRUs) characterized by unique combinations of land cover, soil and slope attributes. In a variety of the previous studies, the interaction between hydrographic features and landscape characteristics was taken for granted in the SWAT model, and calibrated model was frequently used to predict the eco-hydrological variation induced by possible land use changes without investigating sensitivity of model on changes of landscape pattern, and the results of model would be used to provide scientific foundation for decision-making in land use management. For a better understanding of SWAT model response to different landscape patterns, this study obtained a series of SWAT model outputs in Lao Guanhe watershed which is one of the head source of Danjiangkou Reservoir Area, by setting two land use scenarios with different landscape metrics, which are characterized by landscape composition and configurations, and proposed to quantify landscape pattern. By comparing the results of SWAT model using the real land-use data in 2000 to that using the simulated land use data, which have exactly same mean slope and area of each land use type in each sub-basin as the real land use map. It was found that, the SWAT model could display a substantial correlation between the runoff, sediment yields, non-point source pollutants and with slope and area of land use type, but failed to demonstrate the sensitivity of the results of SWAT model to the difference in landscape metrics. Perhaps because the model calculated each HRU separately, then summarized them for the output of the watershed. That is to say, if the SWAT model was applied to evaluate the effects of planting trees in the riparian strips on water quality of a stream, only the increase of forest land area and the descent of forest land slope would be taken in account. However, the interaction between forest land and other land use types, which might be more significant in eco-hydrological dynamics, could not play a part in water quality evaluation by using SWAT model. With further analysis, it is found that the runoff outputs of SWAT model, by using land-use data with different landscape characteristics, could not only match the measured data, but perfectly match each other, just by adjusting model parameters. Therefore, an outwardly reasonable output of SWAT model could be a perfect simulation of real situations, or come from shrewd calibration of model parameters. As a result, the SWAT model is not the default one to simulate the hydrologic process changed by the change of landscape spatial configuration, the sensitivity of SWAT model on changes of landscape pattern can be enhanced by defining slopes with different gradients.