This paper addresses the potential benefit achievable via model-based, predictive control of the AC system in overheated buildings with large glazed fagades. A sick building in Italy (and more specifically a protractedly and intensely sunlit office on the south-exposed, fully-glazed fa ade) was selected as case study. It was monitored over a week in December. The sensor network mapped the indoor climatic conditions, the surface temperature distribution on the glazed facade and the thermodynamic parameters of the air inlet from the 3-speed fan coil unit serving the room (originally in manual mode). The MPC relied on a grey-box model of the room's thermal dynamics, built on a short-term monitoring dataset (4-day-lasting collection in November) and tuned during the first day of testing. Its ergonomic and energetic performance was assessed on the two days of maximum and minimum solar radiation. Considerably high operative temperature persisted all over the period of observation. However, while in free running the room reached peak values of almost 28 degrees C, during the operation of the MPC the setpoint value of 24 degrees C was never exceeded by more than 2 degrees C. Indeed, the offset stayed within +/- 0.5 degrees C for the 95.8% of the time on the sunnier day and for the 60.8% on the cloudier day, while the PMV remained within the 0.5 range for the 85% and the 80% respectively. No local discomfort from vertical gradients or drafts occurred. Furthermore, the energy consumptions were remarkably contained since the fan operated at minimum speed for nearly the 80% of time.