Abstract: In the past, all landslides were mostly taken as the research object in the susceptibility evaluation, but the difference in types of landslides was ignored, leading to the fact that the index weight did not accurately reflect its impact on landslides. In order to accurately predict the landslide disaster, predecessors have put forward a variety of evaluation models: expert scoring, logical regression, neural network, and so on. Those studies have promoted the transformation of landslide susceptibility mapping from qualitative to quantitative. On the basis of previous studies, this paper analyses the feature and mechanism of landslides in the city of Xining, and puts forward landslide susceptibility mapping based on landslide classification. Through field investigation, landslides in the whole area are divided into soil landslides and rock landslides. Based on GIS platform, the geological data are extracted as raster data. Finally, the weighted information value is used to evaluate the vulnerability of the study area. Statistic methods and ROC curve are used to calculate the success rate. The results show that after landslide is classified the success rates are 82.61% and 82.30%, increasing respectively by 10.9% and 5.2%. Areas with poor geological conditions on both sides of the Huangshui River are transformed into highly susceptible areas. It is confirmed that landslide susceptibility mapping based on landslide classification is an effective means for landslide susceptibility mapping.