A column version of the Arctic regional climate system model (ARCSYM) has been developed for testing general circulation model parameterizations in the Arctic. The ARCSYM column model has been employed for a 23-day period in November to simulate conditions over a multiyear ice floe that has been the site of intensive observations as part of the Surface Heat Budget of the Arctic (SHEBA) project. The large-scale tendencies of temperature, moisture, and wind are specified with values obtained from a special column data set obtained from the European Centre for Medium-Range Weather Forecasting. Comparisons between the ARCSYM column simulations and SHEBA data reveal that modeled temperature profiles are too cold aloft and generally too warm in the boundary layer. The occurrence of low clouds is severely underpredicted while the high cloud fraction is over predicted. The modeled longwave radiative cooling at the surface is 1.5–3 times as large as that observed. Much of this bias is related to problems with the treatment of clear-sky radiative transfer and in the simulated cloud optical properties. At the same time, the magnitude of modeled downward sensible heat flux at the surface is much too large. This has been related, in part, to the method for scaling temperature at the lowest modeled level to its surface air value under conditions of strong static stability. The importance of properly treating longwave radiative transfer under extremely cold, clear-sky conditions is evident in the sensitivity studies. The best simulation of cloud properties was achieved by assuming liquid cloud processes and properties at temperatures above 255 K. This temperature is significantly colder than that used in many climate models. The occurrence of supercooled clouds in the simulation dramatically reduced longwave cooling at the surface due to increases in the optical depth and fractional coverage of clouds. Results from a coupled sea ice-atmosphere simulation reveal that improvements in the atmospheric parameterizations are enhanced when the system is coupled.