Abstract: The landslide hazard spatial prediction is to predict the possible spatial location of landslide. Up to now, a lot of spatial prediction methods have been developed combined with GIS mapping at home and abroad. This paper summarizes some commonly used spatial prediction methods and compares their advantages and disadvantages with real applications. It is concluded that deterministic models have clear physical and mechanical meaning, but the assumptions and parameters needed are too many; the statistical methods are relatively simple and used widely, but it's difficult to describe the non-linear relationships between landslide and their related factors; The mode recognition models are suitable for non-linear issues but existed some problems in application such as time consuming, parameter determination and noise influence; the fuzzy comprehensive judging method is more suitable for single landslide assessment. Although the prediction methods become more complex, the prediction research must be carried out on the basis of solid geological analysis. In accuracy verification of these methods, usually the overlap degree between the locations of the existing landslides and the prediction results is chosen as the judging criteria, so we believe that it would be more reasonable to take the unstable slopes and the reactive ancient landslides as study sample. Because the large complex single landslide needs high spatial prediction and assessment accuracy, it is more feasible to establish deterministic model for each single landslide.