As the most energy-intensive economic sector, the building industry offers a significant potential of energy savings. Heating systems are responsible for the most important part of the energy consumed in buildings. The principal function of heating in buildings is to compensate heat losses through ventilation, building’s envelope, and user's activity more widely. Generally, heating systems in buildings are regulated according to schedules with one or more temperature set points defined according to the occupancy of the building. One of the problems of the efficiency of heating systems lies in their control mode which often does not allow to anticipate the possible disturbing events. The conventional control method, which is the most widely used, regulates the heating by studying the response time of the building. However, it is not able to anticipate other phenomena such as meteorological variations (e.g., variation of the outside temperature) and to use the thermal inertia of the building to avoid overconsumption or uncomfortable situations. This paper proposes a data-driven predictive control method for building heating systems in order to improve thermal comfort and energy efficiency. Thereafter, to validate this method, the heating of an experimental building was controlled over a period of 21 days.