Passability maps are cartographic studies that are generally used by commanders in order to plan military operations. Pursuant to standardisation documents, they are developed by marking passable, hardly passable and impassable (GO, SLOW GO and NO GO) areas. This article presents a methodology for the generalisation of passability maps that are created automatically. For this purpose, artificial neural networks (ANN) were used, and, specifically, a multilayer perceptron. Teaching the network consisted in presenting the neural network examples of manual generalisation of source maps. The paper describes the manner of preparing teaching data to train artificial neural networks and their implementation, which leads to the creation of the resulting maps. The maps were generated in multiple input configurations of teaching data, which allowed us to conduct comparisons of the obtained maps. Areas of various levels of passability generalised manually by the operator were compared to maps generated by the ANN. In order to test the consistency of maps, Moran's I spatial autocorrelation coefficient was determined. The conducted tests allowed us to obtain the optimum parameters of the generalisation process. The proposed methodology is fully automated and may be applied to any source data in any chosen area.