Uncertainty of disaster districts is an important character in emergency response, which brings great challenge for facility location decision before disasters. Many kinds of resources are reserved in these facilities in order to satisfy the requirement of affected districts. Careful deployment of facilities can decrease the response time and improve the ability of emergency management. In this paper, scenario analysis is proposed for forecasting disaster districts under uncertainty. In order to make robust decision, a robust p-median model is presented for facility location problem in dealing with the specified percent of affected districts. Since the robust model is NP-hard, an approximation algorithm is designed and theoretical ratio is analyzed. Finally, sensitivity analysis and performance evaluation for the proposed model and algorithm are shown by computational instances.