Several countries are considering extending Daylight Saving Time (DST) in order to conserve energy, and the U.S. will extend DST by one month beginning in 2007. However, projections that these extensions will reduce electricity consumption rely on extrapolations and simulations rather than empirical evidence. This paper, in contrast, examines a quasi-experiment in which parts of Australia extended DST in 2000 to facilitate the Sydney Olympics. Using detailed panel data and a triple differences specification, we show that the extension did not conserve electricity, and that a prominent simulation model overstates electricity savings when it is applied to Australia.