The central task in power station control is to meet the system demand while minimising state and input variations in the plant. In order to achieve good set-point following, the control system must make good use of coupling between the different control loops and if possible use information about future set points to smooth out the control activity. Characteristic generalised predictive control sets out to achieve both these objectives and hence would seem ideally suited to power station control. The presence of some fast and slow dynamics implies that the boiler-turbine system is in essence a stiff problem; the characteristic generalised predictive control (GPC) algorithm is suitably extended to take direct account of this. Simulation results show that the characteristic GPC algorithm (a) provides a systematic means of approaching the boiler-turbine problem and (b) affords some improvements in performance over traditional PID control schemes.