The objectives of this study were to evaluate different root parameters for estimating rill detachment, and to compare and quantify the effectiveness of planted and natural vegetation on soil physical characteristics and in reducing rill detachment and erodibility. Laboratory-concentrated-flow flume tests were performed under a flow discharge of 3 L min−1 and a slope of 15° with intact soil samples. A silt loam soil (Calcic Cambisol, in Loess Plateau in China) was sampled in 2015 from laboratory with planted ryegrass (Lolium perenne L.) and in 2014 from natural vegetation fields under 12 to 36-a restoration. The results indicated that root length density was the best parameter for estimating the effects of roots on rill detachment and erodibility for both planted ryegrass and natural vegetation, with exponential reduction equations. The exponents were more negative for natural vegetation (−0.411) than for planted species (−0.217) because of the existence of more complex root networks under more diverse natural vegetation. Rill erodibility of soil samples with roots decreased exponentially from the baseline erodibility of the root-free soil as RLD increased. Root systems of the restored natural vegetation ameliorated soil characteristics (bulk density, aggregate stability and organic matter content) more profoundly than those of the laboratory tests due to stronger bonding and binding effects, resulting in a lower absolute soil detachment rate for a given root density. There existed a vertical decay distribution pattern of root densities with soil depth for natural vegetation and planted ryegrass, and this distribution of root systems reached a steady state after 12 years natural vegetation restoration. More vegetation types under more different hydraulic conditions should be evaluated in future studies to provide better scientific bases for modelling their impacts on rill detachment and erodibility.