An on-line mathematical approach was used to model the 3-D spatio-temporal temperature distribution in an imperfectly mixed forced ventilated room. A second-order model proved to be a sufficiently good description of the temperature dynamics (R-2 = 0.929) of the system for control purposes. Furthermore, it was possible to fully understand the physical meaning of the second order model structure. Using this model, a model-based predictive controller ( MBPC) was developed for a single input single output (SISO) system. The controller was able to accurately control the mean temperature level of four spatial points in the room, and to robustly react to a random local disturbance signal. The results presented in this paper show that MBPC using data-based mechanistic modelling can be of significant importance in the development of a new generation of climate controllers.