The analysis of the dynamical evolution of soil depth is of crucial importance to the understanding of landform processes in soil-mantled landscapes such as soil production, bedrock weathering, landscape lowering, and sediment (yield). The temporal evolution of the regolith thickness is studied through a stochastic model incorporating the dynamics of both soil production and erosion. Serious and extreme erosive processes (e.g., landslides) are considered as random forcings acting discontinuously and abruptly. The dynamics of soil production is described by a deterministic function of the soil depth, and different hypotheses are considered that express such a dependence. The probability distribution of soil depth is analytically estimated as a function of the relative rates of the processes of soil erosion and bedrock weathering. It is found that for different (and likely) values of the parameters the dynamics can be either transport-limited or weathering-limited. Intermediate conditions correspond to bistable systems having two preferential states that are both explored by the process. The conditions controlling the occurrence of bistable systems are also discussed with respect to the steady state (linear) stability of a companion deterministic model. The existence in the deterministic model of two possible stable states separated by a range of unstable configurations is found to be a condition necessary but not sufficient to the emergence of double modes in the soil depth distribution in the stochastic process suggested by this paper. ¿ 2001 American Geophysical Union