The paper concerns the development of a new adaptive PI controller for use in HVAC systems. The process of HVAC control can be described as a first order plus dead time model. A kind of arithmetic of recursive least squares (RLS) with exponential forgetting combined with model matching of a zero frequency method is adopted to estimate the model's parameters while the system remained in closed loop. Then, a simple tuning formula for a PI controller with robustness based on the estimated parameters was used to adjust the controller's parameters automatically while under closed loop. To evaluate the effectiveness of the adaptive PI controller, the proposed method was compared with a H-infinity adaptive PI controller. The simulation results show that the new adaptive PI controller has superior performance to that of the H-infinity adaptive PI controller. (c) 2006 Elsevier Ltd. All rights reserved.