Various landslide susceptibility models can be available in the literature, and each model has its unique advantages and limitations. Previous studies have shown that no single model performs best across diverse geoenvironmental conditions. To seek better prediction accuracy and reliability, this study proposes three different ensemble methods to take advantage of multiple landslide susceptibility models: qualitative matrix ensemble method, semi-quantitative partition ensemble method, and quantitative probability-weighted ensemble method. To illustrate the effectiveness of the three ensemble methods proposed, a case study is carried out in Fengjie County in the Three Gorges Reservoir Region, China. First, with the support of geographic information system, a total of 1550 historical landslides and the associated 12 conditioning factors are compiled, which are used for the training and validation of four selected single landslide susceptibility models, including two statistical approaches (i.e. frequency ratio and fuzzy assessment) and two machine learning approaches (i.e. backpropagation neural network and support vector machine). Then, the three ensemble methods are applied to integrate the outcomes of the four single models. Finally, an extensive comparative analysis is performed between the ensemble methods and single models using the receiver operating characteristics curve and information entropy. The results demonstrate that all the three ensemble methods achieve higher overall prediction accuracy (> 80%) than the four single models (< 80%), and the matrix ensemble method provides the best improvement. Besides, the ensemble methods can also enhance reliability by reducing the statistical discrepancy between distinct single models.