Abstract: The formation and collapse of landslide dam are usually sudden and dangerous. Large floods are likely to form in a short time if dams break,which could cause serious damages to the social and natural environment downstream. Efficient dam stability prediction is an important basis for making emergency response program in advance and reducing disaster losses. In order to estimate the landslide dam stability quickly,this paper chooses 674 groups of cases,and the one-way analysis of variance and the ROC curve for seven basic parameters which can be gained easily have been analyzed. Five parameters including volume,width,length,material and height of the dam have been chosen as the most important ones according to these two analyses. Moreover,based on 135 landslide dams and the five important parameters,three kinds of five-parameters-models are established by using logistic regression,fisher discriminant analysis and decision tree. Besides,in order to overcome the limitations of the dam database,a model of k-means is established. By comparing the accuracy and generalization ability,it indicates that the estimation result of decision tree is better than the other three models,and fisher discriminant analysis is slightly better than logistic regression and k-means. Dam’s volume has the most influence on dam stability for the four models,and landslide dams with larger volume and material diameter are more stable,which is similar with the results of the qualitative analysis. Hongsong dam,Muguaping dam and Nakatsu River dam are used to test the four models,the estimated results are consistent with the reality. Additionally,a comprehensive index CEIM for evaluating landslide dam has been proposed,the results shows that decision tree and Z model proposed by Shi Zhenming et al. are the best. Decision tree not only could solve the multilinear problem between the parameters,but also show the process of distinguish,it probably provides a new way to estimate landslide dam stability.