Abstract: The Sangzhi district of Hunan is taken as the research area. Based on the comprehensive analysis of rainfall and landslide data of the past 30 years in this area, the correlation and partial correlation analysis are carried out between the cumulative rainfall factors, the rainfall factors and the occurrence or not of landslides and the number of landslides. The optimal effective rainfall attenuation coefficient is determined as 0.8. According to the volume and scale of rainfall monitoring data and historical landslide information, the scattered plot of effective rainfall intensity (I) and duration (D) is obtained. The effective rainfall intensity threshold of the landslide induced by each probability is determined, and the classification of landslide hazard grade is carried out. Using the binary data, the logistic regression of the sample data is conducted, and the disaster probability prediction equation of the study area is obtained. The expression of the critical value of the rainfall intensity induced by the landslide in the study area is also obtained. The rainfall landslide event and the rainfall landslide event are selected for verification. The results show that the landslide prediction model established in this paper is of high accuracy and can provide a scientific basis for rainfall landslide prediction.