Over the past one and a half decades developments in our understanding of earth physics and the increase in computer power have enabled quantitative modelling of subsurface temperature, hydrocarbon charge, migration, overpressures and palinspastics. The North Sea has provided the test bed for many of these efforts. Initial developments focused on one dimensional models of temperature and charge as these were computationally possible at that time, and by the late 80s such models were widely available. Despite the current availability of 2D charge modelling tools, most modelling carried out at present is still ID. Hydrocarbon generation is driven by the temperature field which depends on the heat flow, which is only truly vertical in exceptional circumstances. In principle, a whole lithosphere ID model will only agree with a 3D model for a horizontally layered section. Thermal conductivity anisotropy (lateral/vertical conductivity) of mudstones can reach three. Lateral differences in geology also promote changes in geothermal gradients and therefore lateral heat movement. At the lithosphere scale, differences in the depth to the convecting asthenosphere, and the distribution of hot spots may also cause lateral heat flow. In the North Sea, the complex geology particularly within the syn-rift section makes heat flow strongly 3D. In the Inner Moray Firth and Halten Terrace there are considerable differences between the predictions of fully transient 1D and 3D models for the temperature and hydrocarbon expulsion flux histories of the kitchen areas. Differences can exceed 15°C close to major faults: kitchen areas in the hanging walls are typically cooler than would be expected from ID models, whereas basement cored highs are typically hotter than might be expected. The differences between the predictions of ID and 3D models are driven by structural effects and are thus temporally variant. The importance of 3D effects in charge modelling is compounded in hydrocarbon migration models. Temporal changes in carrier bed interconnectivity, fault seal properties and basin structure influence migration, re-migration and even refilling of breached traps. For instance, hydrocarbons found in the Tertiary of the Central North Sea are derived from underlying Jurassic reservoirs where high overpressures have lead to hydrofracturing and hence leakage. Numerical models of overpressure generation, including rapid sedimentation (under-compaction), thermal expansion (aquathermal pressuring), smectite to illite transition (water generation), etc., tell only part of the story: in order for pressures to rise the fluids have to be contained by rocks with sufficiently low permeability. Furthermore, overpressures are transient and gradually diminish when generation ceases. These processes can be described by a relatively simple differential equation, which can be solved to predict overpressures in one or more dimensions. However, in the North Sea, widespread aquifers (e.g. the Fulmar Formation sands) allow pressure transmission out of synclines. Thus, simple ID models are of limited use, and pressure modelling requires prediction of the stratigraphy and temporal variations in aquifer connectivity resulting from structural movement. Palinspastic restoration is thus a prerequisite. In areas such as the Central North Sea, where there is good pressure calibration and relatively simple stratigraphy, modelling enables prediction of overpressure, trap integrity and reservoir continuity. One problem with physical models discussed above has been their inability to incorporate uncertainties. New models are capable of taking into account uncertainties in both model parameters (e.g. compaction coefficients) and input data (e.g. stratigraphic age) to calculate the uncertainty in the modelled variable (e.g. hydrocarbon flux history). Such models provide a much more realistic evaluation of wildcat prospects. The potential of basin models for prospect evaluation and field development is still in its infancy. From a commercial view point, in areas such as the North Sea where data are readily available to most competitors there is little advantage in having the same models as everybody else. The advantage lies in being able to evaluate acreage faster and with greater accuracy, both in the prediction and the uncertainties.