The aim of this study is to compare the predictive strenghtness of different diagnostic areas in determining landslide susceptibility using frequency ratio (FR), statistical index (SI), and analytic hierarchy process (AHP) models in a catchment from the northeastern part of Romania. Scarps (point), landslide areas (polygon), and middle of the landslide (point) have been tested and checked in regards to their performance. The three statistical models have been employed to assess the landslide susceptibility using eleven conditioning factors (slope angle, elevation, curvature, lithology, precipitations, land use, topographic wetness index (TWI), landforms, aspect, plan curvature and distance to river). The three models were validated using the receiver operating characteristic (ROC) curves and the seed cell area index (SCAI) methods. The predictive capability of each model was established from the area under the curve (AUC), for FR, SI and AHP; the values are 0.75, 0.81 and 0.78 (using polygon as diagnostic area), respectively. Among the three methods used, SI had a better predictability. When it comes to the predictability values regarding the diagnostic areas, the landslide area (polygon) proves to have the highest values. This results from the entire surface of the landslide being taken into account when validating the data. Approximately 70% of the Neolithic sites are located in areas with high and very high susceptibility to landslides, meaning that they are in danger of being destroyed in the future. The final susceptibility maps are useful in hazard mitigation, risk reduction, a sustainable land use planning, evaluation of cultural heritage integrity, and to highlight the most endangered sites that are likely to be destroyed in the future.