In many parts of the world, landslide susceptibility remains inadequately mapped, due to the lack of both data and suitable methods for widespread implementation. Iran is one of those countries with extensive landslide problems, with nearly 4900 large landslides occurring between 1993 and 2007. At the same time landslide susceptibility has not been assessed for the country. Random forest (RF) has recently been shown to be a suitable tool for such mapping. In this study we further coupled the RF method with an evidential belief function (EBF) approach, and tested the suitability for landslide susceptibility mapping for variable terrain and data conditions in the west of Mazandaran Province, northern Iran. Locations of earlier landslides were identified by interpreting aerial photographs and through extensive field surveys. Eleven conditioning factors were used in the RF model. The spatial relationship between landslide occurrence and conditioning factors was then assessed using the data-driven EBF model, and EBF values paired to each map. Finally, the EBF maps were used for running the RF model. Finally, the efficiency of the RF-EBF model was tested using the area under the curve to measure the success and prediction rates of the incorporated data. This resulted in a success rate of 85.2 %, and a prediction rate of 81.8 %. The most important conditioning factors identified were lithology, altitude, distance from roads, and land use, respectively. Based on the overall assessment, the combined RF and EBF approach was found to be objective and an applicable estimator that improves the predictive accuracy and controls for overfitting, and thus useful for landslide susceptibility mapping at regional scales.