Modern rolling stock are equipped with on-board diagnostic system that continuously monitors the operation of critical systems and generates event data when abnormal operation or a specific event is observed. This diagnostic information is typically used retrospectively when the vehicle is called for maintenance, however if the data can be accessed more frequently it could feasibly be used to predict an impending failure with sufficient response time to allow remedial action to be planned before the event becomes terminal. Diagnostic data from a range of vehicle systems has been analysed using different statistical and machine learning techniques to extract trends and features in the data which can then be used in the development of advanced predictive methods for intelligent scheduling of maintenance. Initial outputs from the application of these techniques to diagnostic data showed some encouraging results and potential benefits to support the prediction and optimisation of maintenance within a condition-based maintenance system. Further work is required to improve the prediction algorithms and to integrate with maintenance scheduling tools.