This paper proposed an adaptive neural network controller to maintain the chilled water temperature in heating, ventilation and air conditioning(HVAC) systems. The heat transfer behavior between chilled water and refrigerant is highly nonlinear. It is significant to design a controller that can handle the nonlinearity of the process. Firstly,by analyzing the heat transfer process from mechanism perspective, factors which influence the process have been obtained. Then the frequency of the compressor is manipulated to control the chilled water temperature in the outlet of the evaporator and uncontrolled variables are taken into the neural network controller. With a novel adaptive law for the neural network controller, both the nonlinear phenomenon and disturbance of uncontrolled variables can be handled. To further illustrate the performance of the NN controller, experiment was conducted on a pilot HVAC system. Then the result was compared with that of conventional PID controller. Real time experiment result showed the effectiveness of the adaptive neural network controller.