The aim of the study is to assess the degree of spatial agreement among different patterns of landslide susceptibility maps with almost similar success and prediction rate curves, obtained using different combinations of predictive factors. Our approach was tested in an alpine environment (Italian Alps) where debris flows represent one of the most frequent dangerous processes. A data-driven Bayesian method (the Weights of Evidence modelling technique) was successfully applied, and success and prediction rate curves were computed for supporting the modelling results and assessing the robustness of the models. The values of the area under curves were very similar for different models, ranging from 84.36 % to 86.49% for the success rate curves and from 82.46% to 85.66% for the prediction rate curves. Then, the post-probability output maps were classified into rank-based maps, by using an equal-area criterion, to compare the predicted results. Afterwards, appropriate statistical techniques (Kappa statistic, Principal Component Analysis, and Distance Weighted Entropy) were applied. Kappa statistic and Principal Component Analysis outcomes called for significant differences within the output spatial patterns of the predicted maps as well as within the highest susceptibility classes. Moreover, the estimated Distance Weighted Entropy values showed a very low overall entropy at the valley bottom, as all models predicted this area equally as low susceptible. In contrast, areas characterised by the highest values of entropy were more concentrated in the northern and southern slopes of the study site, lying in zones where landslide density was higher. Consequently, susceptibility maps with similar predictive power may not have the same meaning in terms of spatial pattern of predicted results. It is for this reason that landslide susceptibility maps should be distributed together with map documents aimed at defining the level of accuracy of the predicted results to provide the end-users with informative selection criteria.