Simulation of soil temperatures under forest and range fires requires reliable estimates of soil thermal properties over a range of temperatures from ambient to about 600°C. We measured thermal conductivity of soil samples differing in texture, bulk density, water content, and temperature and then fit the results with a modification of the de Vries equation. Thermal conductivity increases dramatically with temperature in moist soil, reaching values 3 to 5 times the ambient value at 90°C. The modified de Vries theory agreed well with data at low temperature, and provided an upper boundary for measurements at higher temperatures. Only four parameters are required to specify thermal conductivity as a function of bulk density, temperature, and water content: thermal conductivity of the mineral fraction, water content at which liquid flow becomes limiting, a power for the liquid flow function, and a shape factor. Adequate estimates of the conductivity of the mineral fraction can be obtained from handbooks if the soil mineralogy is known, and the water content for liquid return flow can be predicted from soil texture. The other two parameters show a fairly narrow range of variation and can probably be estimated with sufficient accuracy for most simulation purposes.