Landslide boundaries and their spatial shapes are usually presented as irregular polygonal surfaces such as semicircles and bumps, but some landslide susceptibility modelling uses idealized landslide points or buffer circles as landslide boundaries, bringing the uncertainty to susceptibility modelling. To study the in uence of different landslide boundaries on modelling uncertainty, 362 landslides and 11 environmental factors in Ruijin City were selected, and established landslide boundaries and their frequency ratio correlations with environmental factors based on landslide points (Point), buffer circles (Circle) and accurately decoded and drawn polygons (Polygon), respectively. Afterwards, Deep Belief Network (DBN) and Random Forest (RF) were selected to construct models such as Point, Circle, Polygon-based DBN and RF. Finally, the modelling uncertainty analysis was carried out using receiver operating characteristic (ROC) accuracy, the distribution pattern of the susceptibility index and its variability.The results indicate that: (1) Using landslide points or buffer circles as landslide boundaries will increase modelling uncertainty, while using accurate landslide polygon boundaries is more effective in ensuring modelling accuracy and reliability. (2) The uncertainty pattern of landslide susceptibility modelling based on DBN and RF is the same, but the prediction accuracy of the RF model is lower than DBN, moreover, its uncertainty is higher than DBN. (3) The landslide susceptibility results obtained by using points and buffer circles as landslide boundaries can re ect the spatial distribution pattern of landslide probability in the study area as a whole and can use as an alternative solution in the absence of accurate landslide boundaries.