The experiment was conducted in early August in the winter pastures of Sunan deer farm, Sunan Yugur Autonomous County, Gansu Province. Through observation of grazing behavior and the growth of vegetation, the main feeding path of Cervus elaphus kansuensis grazing is confirmed, and the grazing rate from high to low is formed from the import and export of the pasture. Three main paths were selected to set the sample plots at the entrance of 0 m, 300 m, 600 m, 900 m, 1200 m, and 1500 m, respectively. Each sample area is about 1 hm². According to the calculation formula of stocking rate, the grazing rates were 6.90, 4.85, 3.45, 2.45, 1.45, and 1.00 AUM/hm², respectively. Ten continuous quadrats with 0.25m² were set up for pattern analysis, and a 4 m² quadrat was used for fractal analysis. Using the deviation coefficient and the t-test method, Ripley's K function and Monte Carlo stochastic simulation were used to analyze the pattern of the population of Stipa purpurea and Achnatherum inebrians, and the fractal statistics of box counting dimension and information dimension were used to analyze the population of the dominant plants including S. purpurea and A. inebrians in the grazing system. The pattern analysis showed that there were 5 changes in the population pattern of pasture plants. (1) A part of the plant population showed a cluster distribution pattern under all the herding rates. (2) Another part of the plant population showed a random distribution pattern under all herding rates. (3) The distribution pattern gradually shifted from random to cluster distribution as the stocking rate increased; (4) the distribution pattern gradually shifted from cluster distribution to random distribution as the stocking rate increased. (5) The population cluster distribution and the random distribution appeared alternately on the grazing gradient. The deviation coefficient of S. purpurea population was the lowest when grazing rate was 3.64AUM/hm². The coefficient of variation of the population of A. inebrians was the highest when the grazing rate was 4.16AUM/hm². In the scale of 0-1 m, the results of Ripley's K function of S. purpurea population and A. inebrians population were in random distribution pattern in Monte Carlo stochastic simulation interval. The fractal analysis showed that with the increase of grazing rate, the box counting dimension interval of the population of S. purpurea and A. inebrians were[1.596,1.962] and[1.831,1.945], respectively. The information dimension interval of the population of S. purpurea and A. inebrians were[1.590,1.899] and[1.633,1.913], respectively. When the grazing rate was 4.34 AUM/hm², the difference between the population space of S. purpurea and the population of A. inebrians was the largest and the information dimension was relatively low. At the same time, the population distribution was relatively uniform while the change of the pattern was relatively slow. The economic and ecological benefits were balanced, which was the most suitable for the management of pasture.