Prediction is found to be a part of many more complex reasoning problems, e.g. state estimation, planning and diagnosis. In spite of this, the prediction problem is rarely studied on its own. Yet there appears to be a wide range of choices for the design of the central component in a solution to this problem, the predictive model. We examine some of the alternatives and, as a case study, present two different solutions to a specific prediction problem that we have encountered in the WITAS UAV project.