Fossils demonstrate that deciduous forests covered the polar regions for much of the past 250 million years when the climate was warm and atmospheric CO2 high. But the evolutionary significance of their deciduous character has remained a matter of conjecture for almost a century. The leading hypothesis argues that it was an adaptation to photoperiod, allowing the avoidance of carbon losses by respiration from a canopy of leaves unable to photosynthesize in the darkness of warm polar winters. Here we test this proposal with experiments using 'living fossil' tree species grown in a simulated polar climate with and without CO2 enrichment. We show that the quantity of carbon lost annually by shedding a deciduous canopy is significantly greater than that lost by evergreen trees through wintertime respiration and leaf litter production, irrespective of growth CO2 concentration. Scaling up our experimental observations indicates that the greater expense of being deciduous persists in mature forests, even up to latitudes of 83 degrees N, where the duration of the polar winter exceeds five months. We therefore reject the carbon-loss hypothesis as an explanation for the deciduous nature of polar forests.