This paper aims at predicting the geographic distribution and size of gullies across a landscape using geographic information system (GIS) and terrain analysis. Eleven primary (elevation, upslope contributing area, aspect, slope, three curvature measures, flow direction, flow width, flow path length and “d(As)/ds”) and three secondary (steady-state and quasi-dynamic topographic wetness indices and sediment transport capacity index) topographic variables were produced; together with other digital data collected from other sources (soil, geology) were used to explain statistically the predictive gully erosion map. Three tree-based regression models have been performed based on (1) all variables, (2) primary topographic variables only and (3) pairs of all variables. The highest predictive power regression tree model that has been found combines sediment transport capacity and steady-state wetness indices explaining 80% of the field measurements in gully size. This model proves to be simple, quick, realistic, practical, and can be applied to other areas, thus constituting a tool to help implementing a plan for soil conservation and its sustainable management.