Abstract: To alleviate the insufficient spatial representativeness and accuracy uncertainty caused by the reliance on non-landslide samples in traditional landslide susceptibility assessment. This study proposes a non-landslide-sample-free FR-K-means susceptibility evaluation method, with the Random Forest (RF) model adopted for comparison. The Frequency Ratio (FR) model is used to quantify the conditional probabilities between key environmental factors and landslide occurrence, thereby constructing an integrated factor dataset. The K-means algorithm is then applied to cluster similar hazard-inducing environmental conditions, and the Natural Breaks method is employed to delineate landslide susceptibility zones.Using Pingwu County as the study area and Tongjiang County as the validation area in Sichuan Province, the results indicate that the FR-K-means method outperforms the RF model in capturing landslides within very-high-susceptibility zones, with landslide densities of 0.124 and 0.098 events/km², respectively, compared with 0.112 and 0.092 events/km² for RF. In addition, the proposed method clearly delineates high-risk belts along both sides of river basins and effectively reduces the misclassification of extremely low-susceptibility areas within basin interiors as high-susceptibility zones, thereby decreasing false identification in non-hazardous regions. The results demonstrate that the FR-K-means method exhibits favorable spatial focusing capability and regional transferability, providing a complementary approach for rapid landslide risk mapping and refined management in complex geological environments.