Grazing is one of the major anthropogenic driving factors influencing community structure and ecological function of grasslands. Fencing has been proved to be one of the main measures for rehabilitating degraded grasslands in northwestern China. However, data from combined empirical studies on the effects of different management regimes in desert grasslands are lacking. So we selected long‐term fencing (fenced since 1991), mid‐term fencing and seasonal fencing (fenced since 2002), and adjacent free‐grazing grasslands to investigate vegetation and soil properties on southwest Mu Us desert. Our results showed that fencing increased plant cover, height, aboveground biomass (AGB) of different plant life‐form groups, Shannon–Wiener diversity index, Evenness index, Simpson index, total soil nitrogen, total soil phosphorus, and soil organic matter, but decreased plant density, species richness, Richness index, soil bulk density, water content, and pH. However, 22–24 years of long‐term complete fencing might cause redegradation of vegetation and soil nutrients, characterized by the reduction of some vegetation properties, biodiversity, total AGB, and some soil properties. Seasonal fencing with 11–13 year was more beneficial to vegetation restoration than that with completely fencing measures. Our study suggests that appropriate artificial disturbances, such as seasonal fencing (winter grazing and summer fencing), should be used after long‐term fencing in order to maintain grassland productivity and biodiversity. These findings will help to provide theoretical support for vegetation restoration and sustainable management in grassland under grazing prohibition at Mu Us desert.