[Objective] The impact of different evaluation units （grid units and slope units） on the accuracy and reliability of regional landslide susceptibility assessment was explored in order to provide a scientific basis for landslide risk management and disaster prevention and mitigation planning. [Methods] Taking the Huangshui River basin in Ledu District， Qinghai Province as the study area， 12 influencing factors such as slope， aspect， and terrain relief were selected to construct a geospatial database. The random forest model was used to establish landslide susceptibility assessment models based on grid units and slope units， respectively， with parameters optimized through grid search. The prediction accuracy， factor importance， and susceptibility zoning effects of the two unit-based models were compared using the confusion matrix， ROC curve， and landslide frequency ratio analysis. [Results] ① Annual rainfall was the primary controlling factor for both unit types， but the importance ranking of the remaining factors showed significant differences， reflecting the scale effect of spatial division methods. ② The random forest model performed well under both unit types （slope unit AUC=0.905， grid unit AUC=0.838）， with slope units outperforming grid units in indicators such as accuracy， recall， and F₁ score. ③ The susceptibility zoning results indicated that grid units exhibited stronger clustering of disaster points in high-risk areas， making them more suitable for engineering management or detailed planning. Slope units achieved better overall accuracy， facilitating regional management. [Conclusion] Slope units show greater advantages in overall model accuracy， making them suitable for regional disaster prevention management， while grid units perform better in fine-scale assessment of high-risk areas， making them applicable for engineering management， and capable of promoting the refined implementation of disaster prevention and mitigation planning.