This paper describes the results of a preliminary investigation regarding an agent-based methodology for optimizing building HVAC system performance. A typical commercial building HVAC was modeled and its operation was simulated using standard models and a commercially available programming environment. The effect of varying chilled water and supply air temperatures on total system power requirements for a variable-air-volume (VAV) cooling system with a single chiller, and associated pumps and fans was determined. The operation of the agents and the implementation of the optimization method were also simulated. Application of the method to real systems is discussed, along with the need for future refinements and extensions.