Abstract: Snow disaster become a gradually increasing trend, due to the global climate change, particularly on the frequency, intensity, and duration in high risk area of extreme weather. In addition, the prevention and reduction of natural disaster have posed a major challenge on the international community. Therefore, it is urgent to accurately predict the disaster risk, in order to reduce the losses from the snow disaster. Being one of the most important bases for stockbreeding, the Qinghai-Tibetan Plateau is a typical region with high incidence of snow disasters in western China. In the high risk area of snow disaster, forage reserves play a vital role in the emergency response to snow disasters in pastoral areas. The objectives of this study were, 1) to evaluate the high risk area of snow disaster, and to verify the accuracy using the actual occurrence areas; 2) to propose a new method for the site selection of forage reserves by modifying P-median model; 3) to explore the optimum sites for forage reserves over the Qinghai-Tibetan Plateau. The high-risk areas of snow disaster and the occurrence areas in actual years were combined, in order to better evaluate the risk of snow disaster in disaster risk management, and thereby to improve the disaster resilience of plateau pastoral areas. The number of livestock in county-level administrative units and vector data of national basic roads were collected from government units and the National Basic Geographic Information Center. The P-median model was selected to evaluate the sites selection, compared with other models, such as P-center model, maximum coverage model, P-median model. In addition, a modified P-median model was proposed to evaluate the optimum sites of forage reserves in this study area, according to the calculation demand for livestock forage, and the optimal path from the facility to the demand of models. The high-risk area of snow disaster for the Qinghai Province can serve as a case to verify the models. The results showed that 1) Zhiduo, Maqin and Dulan counties were the optimum forage reserves in this study area, which the number of their forage sites was different, such as 7 sites in Zhiduo county (including Maduo, Qiuzhi, Duocai, Zhaduo, Jieduo, Jinisai, and Baizha towns), 4 sites in Maqin (including Longzang, Lajia, Jianshe, and Ningmute towns), and 2 sites in Dulan county (including Xiangride and Jiermeng towns); 2) There was an obvious difference among sizes, where the capacities of Zhiduo, Maqin, and Dulan forage reserves were 210 000, 132 000 and 135 000 ton of forage grass, respectively; 3) In space, a dual structural pattern was formed in “networked layout, hierarchical reserve”, where Zhidu, Maqin and Dulan reserves served as the center, Maduo, Zado, Qiuzhi and other forage reserves as the periphery, covering the entire study area. The finding demonstrated that the P-median model can be expected to perform excellent sites selection of forage reserves in the Qinghai-Tibetan Plateau. The finding can provide an available theoretical method for the selection of forage stock reserve in plateau pastoral area.