In this paper, learning/repetitive control is proposed for improvement of existing feedback control loops for temperature regulation in buildings. A single zone office building is used as an example, with real weather data for Phoenix Arizona and realistic occupancy load schedules. Simulations have shown a decrease in the average set point tracking error of more than 50%, even without additional energy consumption. This can be achieved in situations where the load disturbances have enough repeatability and a repeatable-to-nonrepeatable ratio can be computed to determine if learning should be used and at which frequencies. Furthermore, the increased tightness in reference tracking could be used to lower energy consumption by moving the reference set point closer to the boundaries of the allowable temperature range.