Landslide susceptibility assessment is essential for disaster management. The aim of this work is to select a reliable and accurate model for loess slide susceptibility assessment. We have a frequency ratio model and artificial neural network to develop loess slide susceptibility maps. We analyzed the relationships between loess slide frequency and conditioning factors including elevation, slope gradient, aspect, profile curvature, thickness of loess, rainfall, topographic wetness index, valley depth, distance to rivers, and land use. We developed a landslide inventory consisting of 223 loess slides by the interpretation of remote sensing images from earlier published/unpublished reports and from intensive field surveys. From these 223 loess slides, 178 (80%) were selected for training the models and the remaining 45 (20%) slides were used for validating the developed models. The validation was carried out by using receiver operating characteristic (ROC) curves. From the analysis, it is seen that both the frequency ratio model and artificial neural networks performed equally well, while the frequency ratio method is much easier to apply. The loess slide susceptibility maps can be used for land use planning and risk mitigation purpose in loess terrain.