Abstract: Accurate and efficient information of land use types is of great importance for monitoring and management of land resources. In order to improve the accuracy of land use classification in low mountain and hilly areas, the current study applied random forest with Sentinel-2 images and terrain indicators to classify land use types in Lishi town, Jiangjing, Chongqing. A total of 23 features involved three kind of variables, namely, traditional remote sensing indices, red-edge remote sensing indices and topographic indices were derived from the Sentinel-2 images and digital elevation model. Three models with different inputs were developed and compared. The accurate map of land use types and the relative importance of these indices to the classification were obtained by the best model. The results showed that the values of overall accuracy and Kappa coefficient were 86.54% and 0.800 9 for the model with traditional remote sensing indices, 87.17% and 0.810 2 for the model with traditional remote sensing indices and red-edge remote sensing indices, 88.61% and 0.831 4 for the model with traditional remote sensing indices, red-edge remote sensing indices, and terrain indicators, respectively. The top five importance variables was ranked in order of B2（Blue), B4（Red), B3（Green), Modified Normalized Difference Water Index（MNDWI), and B5（Vegetation Red Edge（1) The random forest with Sentinel-2 images and terrain indicators could be a suitable tool for producing accurate land use information over the study area. The results could provide valuable information for land use classification in the areas with complex topography.