An expert controller for air-conditioning plant is developed using a predictive control approach. The design of the predictive control algorithm is based on prior knowledge of the plant and a rule-based supervisor is used to optimize the control performance. A simple model of the air-conditioning system is derived that has parameters which are easily related to the operating conditions of the process. Safety factors are introduced to account for the uncertainty in the prior knowledge and nonlinearities in the plant. Constraints imposed by the operation of the plant are incorporated directly in the control algorithm. The achievable control performance is limited only by the level of uncertainty and degree of nonlinearity of the plant. Experimental results are presented which show that the use of a rule-based supervisor can lead to significant cost savings without unacceptable increases in discomfort levels. The results abo demonstrate that the expert controller is able to compensate for day-to-day variations in control performance. The development and testing of a dedicated version of the controller is also described.