A classification of HDR models by their representation of the reservoir geometry is presented together with the ＇building-block＇ sub-processes which embody much of the coupled physics necessary for HDR modelling. A number of current modelling approaches are reviewed. The need for integrating stimulation and circulation models is identified. The need for a good, though not necessarily fully explicit, representation of the reservoir fracture geometry is necessary as a vehicle to express the coupled physics. Limitations on the computational tractability of explicit network models, when strongly coupled physics are added, suggests the types of model that may hold the most promise for future work.