This study investigated a highway slope 2 km ahead of the entrance of the Alishan National Forest Recreation Area, at the mileage of 86 km and 950 m of the Alishan Highway, Taiwan. Countermeasures were conducted after a slope failure. Groundwater wells and inclination wells were installed on the slope. The new idea to improve the accuracy of the empirical rainfall-based criteria comes from developing the relations between rainfall and groundwater fluctuation by analyzing the local groundwater elevation and the rainfall data at the Alishan rainfall station. The potential failure surfaces for shallow collapse and deep-seated landslide in the slope and the relationships between the slope stability and the groundwater level were assessed using Geo-Studio. The variation of the groundwater level with the critical state of the slope were obtained. Based on the analysis results, in each kind of potential failure of shallow collapse and landslide, the total cumulative rainfall (ΣR) corresponding to three slope stability states were determined: ①safe, slope is stable, ②dangerous, slope is possible failure (0 ＜ failure possibility ＜ 100%), and ③disaster, slope will failure (failure possibility = 100%). Finally, combine the three slope stability states for the shallow collapse and the deep-seated landslide, a rainfall-based slope failure warning criteria for the test slope on the Alishan Highway is set up to operate in five stages: (1) safety (ΣR ＜ 440 mm), (2) alert (440 mm ≦ΣR ＜ 580 mm), (3) evacuated (580 mm ≦ΣR ＜ 850 mm), (4) disaster (850 mm ≦ΣR ＜ 990 mm), and (5) catastrophe (990 mm ≦ΣR).