Heating, ventilation, and air conditioning (HVAC) systems in large industrial or office buildings are complex systems with many data points that pose enormous challenges to engineering and control. From an engineering viewpoint, the classical, centralized approach of system design has obvious limitations in flexibility regarding later reconfigurations and layout changes during system operation. Moreover, a centralized control strategy is limited in terms of energy efficiency. Therefore, distributed schemes relying on single room controllers rather than a centralized HVAC controller are desirable. Such a scalable system setup will also facilitate engineering and re-engineering, e.g., by using a scalable function block approach. Accordingly, the actual control strategy has to rely on model-based predictive control implemented in the individual room controllers in a distributed manner. As such controllers are resource-limited devices, the models should be simple and computationally efficient, yet sufficiently accurate to represent the thermal behavior of the room together with its surroundings. This paper presents an analysis of various room models and proposes a minimumcomplexity model. Simulation-based comparison with the state of the art demonstrate no significant loss of accuracy while reducing the implementation effort by 47 percent.