Probabilistic methods play an increasingly important role in volcanic hazards forecasts. Here we show that a probability distribution characterized by competing processes provides an excellent statistical fit (>99% confidence) to repose intervals between 75 vulcanian explosions of Soufrière Hills Volcano, Montserrat in September–October, 1997. The excellent fit is explained by a physical model in which there are competing processes operating in the upper volcano conduit on different time scales: pressurization due to rheological stiffening and gas exsolution, and depressurization due to development of permeability and gas escape. Our experience with the Soufrière Hills Volcano eruption sequence suggests that volcanic eruption forecasts are improved by accounting for these different conduit processes explicitly in a single probability model.