Rainfall is the most relevant factor for the triggering of landslides and it is characterized by an extreme variability. Rainfall analysis is the most frequently adopted approach for predicting landslides. But predicting deep-seated landslides by means of rainfall thresholds is sometimes inadequate due to the complexity of such slopes. In this paper, further study of the loading / unloading response ratio theory applied in landslides prediction was carried out. As the aforementioned effect and character of rainfall was concerned, a prediction for the known Xintan landslide was carried out by means of the loading / unloading response ratio method, in which the action of water was regarded as loading/unloading measures. The results show that it is an effective method for landslides prediction, and it has an advantage over conventional rainfall thresholds or time sequences analysis method.