The potential variability in velocity of normally slow-moving landslides has important implications for risk assessment, design, monitoring, and maintenance of infrastructure. A conceptual approach to predict landslide velocity probability distributions using landslide observations, engineering judgment and Markov models is reviewed. Landslide behaviour types that link historical evidence of landslide displacement and mechanisms of movement to probabilistic predictions of future velocity are proposed. Tentative velocity transition matrices are proposed for five landslide behaviour types which yield limiting state probability vectors corresponding to long-term average annual landslide displacements ranging from 1 cm/yr to 1 m/yr. Typical model outputs are provided and potential model applications are discussed.