Aims The forestland understory vegetation reduces concentrated overland flow through infiltration improvement by roots and raindrop interception by surface cover. However, little has been done to quantify the linkages between understory vegetation cover, roots, and channel erosion, and such information can help assessing the role of the reforestation in soil erosion control. In this study, we evaluated the relationships between channel density, root density, and vegetation cover in forested hillslopes of southwestern China. Methods Twelve locations (four slopes and three positions) of forested hillslopes with a wide range of understory degradation due to litterfall extraction and livestock grazing were selected for the study. Channel density as a measure of rill and (small) gully erosion, root density of different diameter classes, and vegetation cover of all types were determined using field measuring, soil coring and the line transect method, respectively. Soil loss rates were estimated using the caesium-137 ( 137 Cs) technique. Results Rills (depth<0.3 m) with a width of 0.05-0.1 m were the dominate channel erosion in all hillslopes with understory-degradation, and small gullies (depth>0.3 m) with a width 0.5-1.0 m were found at the locations of hillslopes with high understory-degradation. Channel density and soil loss rate increased with the increase in understory-degradation in the forested hillslopes. Simple correlation analysis indicated that channel density was negatively correlated with fine root density (diameter< 1 mm and 1-2 mm) and grass and shrub covers, but not with coarse roots (diam. 2-5 mm and 5-10 mm) and mulch and tree covers. The principal component regression revealed fine root density (diam. <1 mm), shrub and grass covers were the most important predictors for channel density in the forested hillslopes. Tree cover, mulch cover and coarse root density were found to have much less influence on channel density. For the model established from this study using principle component regression, vegetation variables could explain 82 % variance of channel density.