Soil erosion on the Loess Plateau of China is effectively controlled due to the implementation of several ecological restoration projects that improve soil properties and reduce soil erodibility. However, few studies have examined the effects of vegetation restoration on soil properties and erodibility of gully head in the gully regions of the Loess Plateau. The objectives of this study were to quantify the effects of vegetation restoration on soil properties and erodibility in this region. Specifically, a control site in a slope cropland and 9 sites in 3 restored land-use types (5 sites in grassland, 3 in woodland and 1 in shrubland) in the Nanxiaohegou watershed of a typical gully region on the Loess Plateau were selected, and soil and root samples were collected to assess soil properties and root characteristics. Soil erodibility factor was calculated by the Erosion Productivity Impact Calculator method. Our results revealed that vegetation restoration increased soil sand content, soil saturated hydraulic conductivity, organic matter content and mean weight diameter of water-stable aggregate but decreased soil silt and clay contents and soil disintegration rate. A significant difference in soil erodibility was observed among different vegetation restoration patterns or land-use types. Compared with cropland, soil erodibility decreased in the restored lands by 3.99% to 21.43%. The restoration patterns of Cleistogenes caespitosa K. and Artemisia sacrorum L. in the grassland showed the lowest soil erodibility and can be considered as the optimal vegetation restoration pattern for improving soil anti-erodibility of the gully heads. Additionally, the negative linear change in soil erodibility for grassland with restoration time was faster than those of woodland and shrubland. Soil erodibility was significantly correlated with soil particle size distribution, soil disintegration rate, soil saturated hydraulic conductivity, water-stable aggregate stability, organic matter content and root characteristics (including root average diameter, root length density, root surface density and root biomass density), but it showed no association with soil bulk density and soil total porosity. These findings indicate that although vegetation destruction is a short-term process, returning the soil erodibility of cropland to the level of grassland, woodland and shrubland is a long-term process (8–50 years).