From landslide mapping it is known that the frequency of landslide occurrence as a function of their magnitude can be described by a power law in many regions. In order to investigate the magnitude distribution of landslides from a theoretical point of view, we present a physically based landslide model combining aspects of slope stability and mass movement. If the long term driving processes (fluvial or tectonic) are integrated, the model shows self-organized criticality (SOC). The results coincide with results obtained from landslide mapping, so that our model suggests that landsliding may be seen as a SOC process. In contrast to other models showing SOC that are mostly based on cellular automata, our model is based on partial differential equations. The results show that SOC is not a fashion of cellular automata, but can also occur in differential equation models. ¿ 1998 American Geophysical Union