Grazing-induced degradation of grasslands is the primary impediment to the socioeconomic development of Inner Mongolia. It affects the entire environment of northern China. Understanding grassland dynamics is necessary for restoration and sustainable management of these degraded ecosystems. The recovery dynamics of a degraded Leymus chinensis (Trin.) Tzvel. grassland after removal of grazing was studied in comparison with its spatial variation along a grazing gradient, using its climax community as a benchmark. The species composition, diversity, and biomass of the grassland vegetation, as well as the attributes (height, density, and individual mass) of major species, were examined on the eight sites along the grazing gradient and in the recovering grassland over 11 years. The spatial pattern of grassland vegetation along the grazing gradient closely reflected its recovery trajectory over time. Both the spatial and the temporal processes exhibited the same shift in species dominance in association with grazing removal or less grazing intensity. Grassland degradation was accompanied by an increase in species density and a decrease in species size; this trend was reversed during recovery. This result suggested that the degraded grassland is highly resilient and that restoration could occur naturally by reducing or excluding grazing animals. However, some differences existed between the spatial and the temporal processes. Species richness was high on the light- or no-grazing sites along the gradient, but varied little during the recovery of the degraded grassland. Species evenness was high under moderate to light grazing along the gradient and was high at the beginning of the recovery period but not at the end. Although standing biomass improved significantly during the recovery period, it did not change significantly along the grazing gradient. These observed discrepancies were related to the intrinsic difference in the spatial versus temporal processes and are discussed together with the advantage/disadvantage of the grazing gradient versus dynamic monitoring methods in grassland dynamics studies.