Soil erosion by water is a major cause of landscape degradation in Mediterranean environments, including Lebanon. This paper proposes a conditional decision-rule interpolation-based model to predict the distribution of multiple erosion processes (i.e. sheet, mass and linear) in a representative area of Lebanon from the measured erosion signs in the field (root exposure, earth pillars, soil etching and drift and linear channels). First, erosion proxies were derived from the structural OASIS classification of Landsat thematic mapper (TM) imageries combined with the addition of several thematic erosion maps (slope gradient, aspect and curvature, drainage density, vegetal cover, soil infiltration and erodibility and rock infiltration/movement) under a geographic information systems (GIS) environment. Second, erosion signs were measured in the field, and interpolated by the statistical moments (means and variance) in the defined erosion proxies, thus producing quantitative erosion maps (t ha À1 ) at a scale of 1:100 000. Seven decision rules were then generated and applied on these maps in order to produce the overall decisive erosion map reflecting all existing erosion processes, that is, equality (ER), dominance (DOR), bimodality (BR), masking (MR), aggravating (AR), dependence (DER) and independence (IR). The produced erosion maps are divided into seven classes ranging between 0 and more than 1Á8 t ha À1 for sheet erosion, and 0 and more than 10Á5 t ha À1 for mass and linear erosion. They are fairly matching with coincidences values equal to 43 per cent (sheet/linear), 48 per cent (sheet/mass) and 49 per cent (linear/mass). The overall accuracies of these maps were estimated to be 76 per cent (sheet erosion), 78 per cent (mass erosion) and 78Á5 per cent (linear erosion). The overall decisive erosion map with 15 classes corresponds well to land management needs. The model used is relatively simple, and may also be applied to other areas. It is particularly useful when GIS database on factors influencing erosion is limited.