Canonical correspondence analysis (CCA) and regression models, developed to predict mean annual temperature (MAT) using leaf morphology and wood anatomy, were tested using leaves and wood collected from sites in the eastern United States, and were used to calculate paleotemperature of the Eocene Clarno Nut Beds. For leaves, regression based on the single character "margins entire" gave the most accurate results, with errors less than 5 degrees centigrade at the modern sites. For wood, CCA using 13 anatomical characters gave the most accurate results, although errors were larger than those for leaves. CCA revealed that, unlike leaf physiognomy, a change in wood physiognomy occurs between temperate and tropical regions, with temperate wood physiognomy more strongly influenced by MAT than tropical wood physiognomy. Therefore the use of wood to compute paleotemperature from fossil assemblages that have a tropical component (as does the Clarno assemblage) is questionable.

Key words: Clarno, leaf morphology, paleotemperature, temperature, wood anatomy