Gas outburst is recognized as a potentially fatal hazard to be managed during the mining of gassy coal seams. Gas outbursts in Australian coal mines have been associated with the presence of geological structures in the coal and surrounding rocks, having a range of spatial scales from millimetres to metres. The contributing mechanisms are influenced by coal gas reservoir variables including gas composition, fluid pressure, desorption pressure and rate, porosity, intrinsic permeability and relative permeability. They are also influenced by geomechanical variables such as strength, in situ stress and mining-induced stress. The work described here concerns the influence of variability of permeability and strength on outburst behaviour. Key coal properties are characterised for a field site in an operational underground coal mine and then used in a series of hypothetical simulations of outburst initiation. At the field site an array of horizontal holes was cored in-seam. Well tests for permeability and stress were performed in the holes and recovered core was tested in the laboratory for permeability, strength and sorption properties. For permeability and strength sufficient numbers of measurements were obtained to describe these properties statistically assuming that the measurements are uncorrelated spatially. Analysis of the cumulative probability distributions of the field and laboratory permeability data shows a strong size effect that cannot be linearly up-scaled. Using Monte Carlo techniques, these statistical descriptions are used to generate realisations of property values across the model grid used in the simulation analyses. The stochastic model approach demonstrates that the variability of the permeability and strength fields can lead to both outburst and non-outburst outcomes from the same measured input data, depending on the corresponding spatial distribution of permeability and strength at the face. These results suggest the potential application of this approach as a tool for outburst risk analysis and assessment. (C) 2008 Elsevier B.V. All rights reserved.