Landslide represents morphodynamic processes that occur in steep terrain and the residential, industrial, gardens and arable land damage. In this study, the zoning landslide in the basin Songurchay Multilayer Perceptron model was used. For this purpose, the data of 124 landslides were identified and presented to the system. In addition to processing landslide in the software MATLAB, 7 layers of layers of slope, aspect, DEM, lithology, hydrographic network, NDVI and the distribution of landslides in Arc GIS software was drawn. Layers based on the normalized value for each layer in the range between 1 and zero respectively. Then 80% of the data for training and 20% for test data selection and the structure of the multilayer perceptron model were analyzed. Due to the nonlinear relationship between the phenomenon of landslides sigmoid transfer function was used. For this study, the number of neurons in the middle layer was changed from 5 to 20 and 18 neuron network with minimal error in the training and testing was. The learning rate of weight change in control, from 0.1 to 0.3 was tested with the 0.3 network to achieve the best learning. Finally, with respect to the weight of the output, landslide zonation map was drawn. The results showed that the geological structure of the mountains made of high potential benefit in the event of landslides.