In a hydronic heating system, it is important to control the zone temperature with a certain degree of accuracy. The system consists of a space heating circuit with the ability to modulate both the fuel mass flow of the burner and the mass flow of hot water. The physical model of the system consists of a heating system, a distribution system and an environmental zone which are simulated by an eleventh order nonlinear system. The objective of the controller is to maintain the zone temperature as close as possible to a chosen setpoint and, at the same time, to maintain the temperature of the hot water below a maximum temperature which is usually 90 degrees C. However, there are many problems in designing such a controller due to system uncertainties, saturation of the actuators and long system delay. To overcome the above nonlinearities, an adaptive controller has been designed using neural networks and evaluated with software simulations. Comparisons are made between the RMS of the zone temperature and the energy consumption of the system for a time period of two days by several neural networks with different architectures and learning laws.