The Urban Dispersion Model (UDM) calculates the time sequence of concentrations, sizes and positions of multiple dispersing puffs in urban areas. For speed and ease of use, each puff is approximated by a 3D gaussian concentration profile. The UDM has been validated against extensive field trial and wind tunnel data, and found to be robust and reliable in application. For a given wind field the UDM produces a single result, but this masks the stochastic variability inherent in physical dispersion experiments. In effect, the UDM calculates the ensemble average of the variability, suitable for the definition of a broad ‘hazard area’, but this cannot be used for the prediction of temporal or spatial fluctuations, which in some cases are critical determinants of risk outcomes. Accordingly, the Probabilistic Urban Dispersion Model (PUDM) has been developed, which, in addition to its UDM functionality, predicts the impact of stochastic variability in atmospheric dispersion. This paper sets out the formulation of the PUDM, and gives examples of its application.