Early warning systems are important measures for sustainable development of risk reduction and livelihood protection. Advantages and improvement have emerged in debris flow warning systems developed in Taiwan and Japan. This study took the Sheng-Mu region, which has frequent debris flow events, as the research area and used the watershed as the warning unit. Both hourly rainfall data and effective accumulated rainfall data are used for issuing warnings. This study established several safety curves with survival analysis theory. The model derives the warning critical line by analyzing the warning hit rate, false alert rate and evaluation values based on none-event rainfall data and debris flow event data. The result of this study showed that the warning critical line can be established quickly on the basis of non-event rainfall data regardless of whether debris flow events exist in a specific area.