The purpose of this research is to develop and compare two machine learning methods, namely, artificial neural network (ANN) and random forest (RF) used for in landslide susceptibility mapping (LSM) of Wushan County. Firstly, 866 landslides were collected from extensive field investigations, historical records and satellite images from 2001 to 2016. Based on previous literature reviews and the field investigations, a geospatial database was established in geographic information system (GIS) based on topography, geological conditions, environmental conditions and human activities factors, including 22 conditioning factors. Then, based on samples of landslides and non-landslides, the 10-fold cross-validation was used to select the best training and test datasets. Finally, the LSMs of Wushan County were generated. Subsequently, susceptibility maps of the two models were divided into: very low, low, medium, high and very high class using experts’ experience method. The two models were compared with area under the receiver operating characteristic (ROC) plot values (AUC) and confusion matrix. The AUC values of the ANN and RF models’ test dataset was 0.966 and 1.000, and the accuracy were 0.953 and 1.000, respectively. According to the mean decrease Gini of RF model, the most important conditioning factors were elevation (a mean decrease accuracy of 72.38), followed by annual average rainfall (47.95) and POI kernel density (45.63). Above all, it is concluded that both ANN and RF models were satisfactory for the LSM of Wushan County at the mountainous region. The RF model had an outperformer prediction and is considerably more efficient than the ANN model.