Abstract: Fine investigation of geological disasters forms the foundation for effective disaster risk mitigation in township-level regions. However, appropriate evaluation models and methodologies that match the scale and required resolution of township-level assessments remain limited. Based on the geological disaster refinement survey in Xiongjia Township, Wanzhou District, rocky landslide hazards were subdivided into dip slopes and non-dip slopes according to the slope structure, and then evaluation index system was established separately. Given the limited availability of historical landslide data at the township scale, an Analytic Hierarchy Process (AHP) and a Random Forest (RF) suitable for small-sample calculations were used for evaluation. The points of unstable slopes judged by the refined survey were included in the landslide samples, and the points of stable slopes were included in the non-landslide samples, which were used for the establishment and testing of the RF model. The results show that the susceptibility evaluation zones based on the two models are generally consistent. The high and very high susceptibility zones of compliant rocky landslides are mainly in the south-central part of the country, related to the soft strata, the proximity conditions, and the human process activities, while the high and very high susceptibility zones of non-compliant rocky landslides are mainly in the northern part of the country, related to the slope gradient, the density of the gully, and the number of movable blocks. The AUC values for AHP-based evaluations were 0.803 (dip slopes) and 0.798 (non-dip slopes), while RF-based evaluations achieved AUCs of 0.849 and 0.876, respectively. Overall, the accuracy of RF was higher than that of AHP, in which the accuracy of AHP evaluation of nonconformal slopes was higher than that of conformal slopes, and the opposite was true for RF. This study provides a basis for local disaster prevention and mitigation and for the ongoing regional geohazard refinement investigation and evaluation in China.