Probabilistic methods provide a means of demonstrating the potential variability in predictions of coastal cliff recession. They form the basis for risk-based land use planning, cliff management and engineering decision-making. A range of probabilistic methods for predicting soft coastal cliff recession has now been developed, including statistical techniques, process-based simulation and structured use of expert judgement. A new episodic stochastic simulation model is introduced, which models the duration between cliff falls as a gamma process and fall size as a log-normal distribution. The method is applied to cliff recession data from a coastal site in the UK using maximum likelihood and Bayesian parameter estimation techniques. The Bayesian parameter estimation method enables expert geomorphological assessment of the local landslide characteristics and measurements of individual cliff falls to be combined with sparse historic records of cliff recession. An episodic simulation model is often preferable to conventional regression models, which are based on assumptions that are seldom consistent with the physical process of cliff recession.