In 1997, we conducted a vegetation survey in three semi-arid natural grasslands (steppes) withdifferent livestock grazing intensities in Southwest Heilongjiang Province, China. The dominant grasslandspecies was the grass Stipa baicalensis Roshev. Grasslands with light, intermediate, and heavy grazingintensities were located 10, 5, and 2 km from a village, respectively. Villagers use the steppe to raise cattle,horses, sheep, and goats. Each of the three grasslands was surveyed by placing 100 quadrats (50 cm×50 cm)along a 50 m line transect. Each quadrat was divided into four equal areas (25 cm×25 cm; S-quadrats) and allplant species occurring in each of these smaller areas were identified and recorded. These data were summa-rized into frequency distributions and the percentage of S-quadrats containing a given species and thevariance of each species were estimated. The power law was applied to these estimates. The power law wasused to evaluate the spatial heterogeneity and frequency of occurrence for each species in the grasslandcommunity. The lightly grazed grassland exhibited high spatial heterogeneity (caused by large plant size),the highest species diversity, and a high occurrence of S. baicalensis. In contrast, the heavily grazedgrassland exhibited high spatial heterogeneity (caused by patchy populations of small plant size), lowspecies diversity, and a low occurrence of S. baicalensis. We judged that the heavily grazed grassland wasovergrazed and exclusion of livestock from the degraded areas is necessary for recovery.