Cloud and boundary layer variables from the European Centre for Medium-Range Weather Forecasts (ECMWF) forecast model were compared with measurements made from surface instruments and from upward looking 8 mm wavelength radar and lidar at the Surface Heat Budget of the Arctic Ocean (SHEBA) ice camp during November and December of 1997. The precipitation accumulation, near-surface winds, and surface downward longwave irradiance predicted by the model were in good agreement with SHEBA observations during this period. However, surface downward longwave irradiance was underestimated by 10 W m-2 on average when low clouds were present in the model and observations. The model demonstrated considerable skill in predicting the occurrence and vertical extent of cloudiness over SHEBA, with some tendency to overestimate the frequency of clouds below 1 km. A synthetic radar reflectivity estimated from the ECMWF model variables was compared with 8 mm wavelength radar measurements. The two were broadly consistent only if the assumed snowflake size distribution over SHEBA had a smaller proportion of large flakes than was found in previous studies at lower latitudes. The ECMWF model assumes a temperature-dependent partitioning of cloud condensate between water and ice. Lidar depolarization measurements at SHEBA indicate that both liquid and ice phase clouds occurred over a wide range of temperatures throughout the winter season, with liquid occurring at temperatures as low as 239 K. A much larger fraction of liquid water clouds was observed than the ECMWF model predicted. The largest discrepancies between the ECMWF model and the observations were in surface temperature (up to 15 K) and turbulent sensible heat fluxes (up to 60 W m-2). These appear to be due at least partially to the ECMWF sea ice model, which did not allow surface temperatures to respond nearly as rapidly to changing atmospheric conditions as was observed.