Understanding of driver behaviour can provide invaluable insight for the design of vehicles and driver assistance systems. Most existing human driver models do not incorporate driver learning or the effect of a driver’s confidence in their predictions of future states, both of which affect human-generated control actions. In this paper, human driver steering control is assessed based on experimental data, then a driver model is proposed which captures driver learning and is capable of reproducing a wide range of human control styles by the selection of appropriate parameters. The driver model learns an internal model of the vehicle dynamics from experience, using a Gaussian Process, then selects control actions via Model Predictive Control. The results verify the model’s capacity to capture the learning drivers achieve over time and to replicate various observed cautious and adventurous behaviours.