The study was conducted in the Caijiachuan watershed in Ji County, Shanxi Province. The primary objective was to quantify streamflow differences among small watersheds with different vegetation covers. Annual runoff coefficient ranged 1.6%-2.3% and 3.1%-3.9 % in the forested and the non-forested watersheds, respectively. Baseflow was negligible in artificial forest watershed, and but baseflow ratio (baseflow/total flow) was higher (1.0%-1.5%) in other forest watersheds. In the non-forested watersheds, base runoff coefficient was highest (2.5%-2.8%). During the rainy season in June-September, total runoff from the artificial forest watershed was highest, about 3.4 times of the natural forest watershed, or 1.9 times of the agricultural watershed. The baseflow in the agricultural watershed was highest, about 2.2 times of the natural forest watershed. For storm events with short duration and high intensity, runoff from the artificial forest watershed was highest, about 10.8 times of agricultural watershed. For storm events with a long duration, peak discharge per unit area in the artificial forest watershed was highest, about as 6.9 times of the natural forest watershed. For large storm events with a long duration, runoff in the agricultural watershed was highest, about 1.8 times of the natural forest watershed. Compare with all landuses and vegetation covers, bench terraces and natural forests could greatly reduce overland flow and had large water storing capacity. We found the artificial forests did not have the hydrologic functions to reduce peakflows. We conclude that natural vegetation recovery had the best hydrologic benefits in soil conservation and thus should be promoted in the Loess Plateau area.