In this paper, we examine the performance of three land surface schemes-the Biosphere-Atmosphere Transfer Scheme (BATS1E), the NCAR Land Surface Model (LSM) and the Canadian Land Surface Scheme, (CLASS) for simulations at two sites on the Alaskan North Slope: Barrow and Imnavait Creek. These sites are characterized by similar large-scale meteorological conditions but different land surface environments and soil characteristics. The simulations are driven by field data collected in several observational programs during the 1992 Arctic summer season; these programs also provide validation data for the schemes. Examination of the simulation results indicates that the schemes with three or more layers (LSM, CLASS) capture the basic seasonal behavior somewhat successfully, while BATS1E is less satisfactory. The details of the soil thermal and moisture field evolution vary substantially between LSM and CLASS. The results suggest that improvements in the treatment of Arctic biological processes, water and energy partitioning, and freeze-thaw physics are needed before higher confidence in prediction from not only the land surface models but also their host climate models is possible for the Arctic regions. ¿ 1998 American Geophysical Union