A stochastic slope stability analysis method is proposed to investigate the short-term stability of unsupported excavation works in a soft clay deposit having spatially variable properties. Spatial variability of undrained shear strength is modelled by a stochastic model that is the sum of a trend component and a fluctuation component. The undrained shear strength trend, which is also spatially variable, is modelled by kriging or a random function. Slope stability analyses are performed on the stochastic soft clay model to investigate the contribution of spatial variability of undrained shear strength to a disagreement among high factors of safety computed from deterministic methods for slopes that have failed. Probabilities of failure as computed from the stochastic analyses give a better assessment of failure potential. Probability of failure values also correlate with time delay before failure. This phenomenon may be related to progressive failure or creep and to pore pressure dissipation with time.Key words: slope stability analysis, failure probability, spatial variability, stochastic modelling, geostatistics, vane tests, sensitive clay.