Abstract: Nanjiang County is located in the red bed regions of the mountainous and hilly region of eastern Sichuan Province, where rainfall-induced landslides frequently occur during the rainy season. This study constructed a database of 82 landslide events from 2016 to 2021. Using rainfall data from the gauges closest to each landslide, four key variables were calculated: cumulative rainfall amount (E), rainfall duration (D), average rainfall intensity (I), and antecedent cumulative rainfall amount (A). To analyze the influence of current rainfall, scatter plots of I-D (intensity-duration), E-D (cumulative rainfall-duration), and E-I (cumulative rainfall-intensity) were generated. For the combined effect of antecedent rainfall (preceding a given period) and current rainfall, scatter plots for E-A5, E-A10, and E-A15 were established. Based on the area clustering method, a four-level early warning model was proposed for the region, and its thresholds were validated using both historical rainfall-induced mass landslide events and recent monitoring data. The results indicate that: (1) landslides in the study area are strongly correlated with the peak rainfall intensity of the day (R), the same day cumulative rainfall (E), and the cumulative rainfall from previous periods (A); (2) threshold models such as lgI-lgD, lgE-lgI, E-A5, E-A10, and E-A15 exhibit good classification performance and are suitable for graded early warning applications. These findings provide empirical support for the development of rainfall-induced landslide early warning systems in red bed mountainous areas of Sichuan Province.