We report an in situ experimental determination of the thermal conductivity of natural sea ice, a parameter required by models of climate and sea ice growth. The results obtained in the experiment differ in detail from the theoretically predicted values used by those models. In particular, we find that experimental results are lower than predicted, due to disorder in the crystalline structure of the ice. The reduction is nearly a factor of two near the top surface of the ice. Deeper in the ice, in contrast, heat flow is enhanced by a contribution from brine convection. ¿ 2001 American Geophysical Union