The identification of eroded areas at catchment scale can be very useful for environmental planning and can help to reduce land degradation and sediment yield. In this paper remote sensing techniques are used to discriminate both eroded areas and erosion risk areas in a badlands landscape developed on Eocene marls, in the Ésera River catchment (Spanish Pyrenees). The spatial distribution, the scarce vegetal cover and the high erosion level let a good visual and digital discrimination of badlands, as opposed to other land covers and surfaces. A maximum likelihood supervised method was used to discriminate heavily eroded areas (badlands) from scarce or densely vegetated lands, by means of the spectral signature of bare soil. The classification distance was used to obtain thresholds for eroded areas and areas at risk. Two error statistics (sensitivity and specificity) where used to determine the most adequate threshold values. The resulting map shows that most areas at risk are located surrounding the badland areas.