There’s a tight relationship between landscape pattern and eco-hydrological processes, especially in semi-arid area. Research on this relationship is one of the most important topics of landscape ecology. Models, taking full advantage of limited data to reveal the change rules or mechanisms and simulate the dynamics of landscape, have become powerful tools for the research on the relationship between landscape pattern and eco-hydrological processes. The Loess Plateau of China is famous for the high-intensitive human activities and serious soil loss. For controlling soil loss, a series of conservation engineering and vegetation restoration have been carried out. The landscape pattern in the Loess Plateau has changed a lot. It’s essential to quantitatively evaluate the eco- hydrological effect of landscape pattern change. Modeling research in the scale of watershed is a effective way to solve the issue. Taking the Yanhe watershed (7725km2) locating in the loess hilly-gully region as the study area, this research was carried out on the basis of SWAT model (Soil and Water Assessment Tool). Combined with field investigation and remote sensing tools, the vegetation heterogeneity was analyzed. In addition, the influence of soil classification and spatial discretization on simulation results was assessed. Integrating the results, the model was improved aiming at the characteristics of the study area. The hydrological effect of soil conservation engineerings and vegetation restoration was evaluated based on the improved model. The vegetation investigation and remote sensing image inversion showed that the vegetation coverage pattern in the Yanhe watershed was fragmented. According to the investigation data, leaf area index (LAI) was influenced not only by environmental factor (precipitation), but also the artificial factors such as planting age and planting density. Irregular artificial factors lead to the fragmented vegetation coverage pattern. Aiming at the fragmented vegetation coverage and complex topography in the Yanhe watershed, the classification method of simulation units was improved. New units called “land type units” were delineated based on land use type, soil type, vegetation condition and slope. The model suitability was improved. Taking the Ganguyi hydrological station as an example, in the simulation for years 1956-1960, Nash-Sufficient coefficient (ENS) and the coefficient of determination (R2) for monthly flow was increased from 0.685 and 0.692 to 0.706 and 0.777, respectively; and that for monthly sediment discharge was increased from 0.427 and 0.630 to 0.654 and 0.682, respectively. Soil classification had important influence on the simulation results. When the loess soil group was divided into one soil genus, the hydrological simulation efficiency of the upstream hydrological stations was low. When the loess soil group was divided into sandy loess soil, loess soil, grey loess soil and chalk loess soil, due to the relatively fully consideration of the soil types heterogeneity, the hydrological simulation efficiency was improved, especially for the upstream hydrological stations. However, due to the wide distribution of the loess soil genus, the model failed to consider the spatial variation of the soil attribute. Hydrological simulation efficiency of Zhaoan hydrological station, controlling high area ratio of upstream loess soil genus, was below the standard (ENS > 0.5, R2 > 0.6). By adjusting the attribute parameters of upstream loess soil genus, this problem was settled. On the basis of land type units delineation, soil type classification and soil parameters adjustment, the hydrological processes for years 1956-1960, 1984-1987 and 2006-2008 were simulated. Conservation engineerings increased greatly after 1970s and altered the hydrological processes. Simulation for 1984-1987 and 2006-2008 appeared deviation. The simulation results were treated as base values, and observations as the actual values taking conservation engineerings into account. The deviation between them was used to evaluate the effect of the engineerings. The results were listed below: 1) Engineerings reduced the runoff in general. Taking the Ganguyi hydrological station as an example, for the existence of the engineerings, runoff was reduced by 24.7% in 1984-1987. This trend was more apparent for the increased engineerings. In 2006-2008, runoff was reduced by 37.3%. 2) Engineerings reduced the runoff and sediment yield in rainy season (June-October). Runoff in rainy season in 1984-1987 of the Ganguyi hydrological station decreased by 40.9%, and sediment yield decreased by 37.7%. This trend was more apparent when the engineerings increased. In 2006-2008, runoff in rainy season decreased by 55.4%, and sediment yield decreased by 75.6%. 3) Engineerings increased the runoff in dry season (November-April). Runoff in dry season in 1984-1987 of the Ganguyi hydrological station increased by 104.5%. This trend was mainly because of the increase baseflow in dry season for the existence of engineerings. When the engineerings increased, the intercept effects exceeded the increase effect on baseflow, and the increasing rate of runoff in dry season reduced. In 2006-2008, runoff in dry season of the Ganguyi hydrological station increased by 46.0%. From 1990 to 2008, vegetation in the Yanhe watershed changed apparently. Farmland reduced sharply and grassland increase greatly. Shrubland and forest increased slightly as well. The hydrological effect of vegetation change between 1990 and 2008 was assessed based on model. The results showed that runoff and sediment yield was reduced owing to the vegetation change. Runoff of the Ganguyi hydrological station reduced by 4.7%, and sediment yield reduced by 11.2%. The hydrological effect of vegetation change was different for different hydrological stations and different climate. In general, when cropland decreased, forestland increased and high coverage grassland increased, runoff and sediment yield decreased more obviously. The less precipitation, the more obvious decreased trend runoff and sediment yield performed.