The paper presents an adaptive optimum start-up time controller and an adaptive optimum shut-down time controller for heating systems based on a robust gradient method. Using very simple models of the heating and cooling processes, it is concluded that for each controller only one parameter is necessary to identify the correct values of the preheat and shut-down time. A robust gradient method is also derived for the tuning of the unkown parameter of the controllers. The behaviour of the algorithms is investgated by implementing them in a laboratory test-bed and in a real building. Results are presented which show the good performance characteristics of the controllers, the important energy savings made when they are applied to real buildings, and their suitability for low-cost microcomputer implementation.