Automatic ground filtering for Light Detection And Ranging (LIDAR) data is a critical process for Digital Terrain Model (DTM) and three-dimensional urban model generation. Although researchers have developed many methods to separate bare ground from other urban features, the problem has not been fully solved due to the similar characteristics possessed by ground and non-ground objects, especially on abrupt surfaces. Current methods can be grouped into two major categories: neighborhood-based approaches and directional filtering. In this study, following the direction of the second branch, we propose a new Multi-directional Ground Filtering (MGF) algorithm to incorporate a two-dimensional neighborhood in the directional scanning so as to prevent the errors introduced by the sensitivity to directions. Besides this, the MGF algorithm explores the utility of identifying pattern varieties in different directions across an image. The authors conducted a comprehensive test of the performance on fifteen study sites and compared our results to eight other publicized methods based on the Kappa coefficients calculated from the error matrices reported by ISPRS. Overall, the MGF filter produces a promising performance in both urban and forest areas. The size and shape of non-ground objects do not pose significant influence on the performance of the MGF algorithm. The fact that MGF algorithm is robust to two commonly required parameters, slope and elevation difference thresholds, has added practical merits to be adopted in different landscapes