The recovery of bitumen (heavy crude oil) from Alberta oil sands depends critically on the coalescence of the dispersed oil drops. When suspended in water, the surfaces of bitumen droplets are known to be negatively charged. According to classical DLVO theory, the resulting double layer repulsion should be sufficiently strong to prevent any coalescence of the droplets. From experience, however, it is known that the oil droplets do coalesce, albeit randomly. Further, the probability of coalescence is seen to increase sharply with bigger drop size. The underlying mechanism of the coalescence process appears therefore not to be deterministic; it may, however, be understood through a probabilistic approach. In this presentation, we propose a model of heterogeneous surfaces which considers the bitumen-water interface to be randomly charged. The DLVO theory, according to this model, remains sound - but only on the microscopic scale. With this new approach, we are able to successfully predict the coalescence probabilities of bitumen droplets as the drop size, degree of drop deformation, and ionic strength are varied. From our experimental results, it is suggested that the bitumen-water interface is composed of patchy domains with a characteristic size of about 0.5 micrometre.