Default options have been shown to affect behaviour in a variety of economic choice tasks, including health care and retirement savings. Less research has tested whether defaults affect behaviour in the domain of energy efficiency. This study uses data from a randomized controlled experiment in which the default settings on office thermostats in an OECD office building were manipulated during the winter heating season, and employees’ chosen thermostat setting observed over a 6 week period. Using difference-in-differences, panel, and censored regression models (to control for maximum allowable thermostat settings), we find that a 1°C decrease in the default caused a reduction in the chosen setting by 0.38°C on average. Sixty-five percent of this effect could be attributed to office occupant behaviour (p-value=0.044). The difference-in-differences model shows that small decreases in the default (1°) led to a greater reduction in chosen settings than large decreases (2°). We also find that office occupants who are more apt to adjust their thermostats prior to the intervention were less susceptible to the default. We find no evidence that offices with multiple occupants displayed different patterns in thermostat choices than single-occupant offices. We conclude that this kind of intervention can increase building-level energy efficiency, and discuss potential explanations and broader policy implications of our findings.