A critical component of any new tunnel design in a complex geological setting is the development of an integrated three-dimensional (3D) geology model with a full set of geotechnical properties of the rocks present. A novel means of predicting the geology based upon sparse observations of structural geology at the surface and from drill holes has been developed and used in several major tunnel studies in Europe.An example of this is the Lyons-Turin twin rail tunnel which is still in a development phase. This tunnel has been in planning and development for over 15 years already. An important aspect of this work is to characterise geological uncertainty along various possible tunnel paths as an input into early design optimisations for time, cost and risk avoidance strategies. Once the tunnel works commence and more detailed geological data is collected by daily mapping, progressive refi nement of the geological model is desirable; geological conditions still to be encountered can be more accurately predicted. Faults and their offsets, dykes, geological unconformities, folding, as well as standard sedimentary processes are all correctly modelled in the software. It is important to allow for seamless interchange of geological models at various scales with CAD workflows so that sufficient detail can be readily accessed at the required scale. Also, the proposed tunnel in its correct geometry can become a section' in its own right. Progressive detailed local geology models in the vicinity of the advancing head are created from the regional scale work by cutting and pasting a starting model, then adding all the extra detail for the required engineering geology work.For all rocks, most standard geotechnical quantities such as sonic velocities, Young's modulus and Poisson's ratio are handled. These and any other phenomena from borehole observations can have variograms calculated and stratigraphically constrained estimations made in a least biased geostatistical manner, in advance of the tunnel. RQD, or a prediction of rock fractures, is handled by the faulting and folding modelling facilities. This is probably the more interesting issue to pursue, as the fractal nature from small scale fractures and joints, to large-scale faults, can now be characterised more convincingly.