Ice-ocean models are being used to understand the causes and mechanisms of ice ocean variability in the Arctic. They are also used to improve physical parameterizations with the ultimate goal of better representation of ice ocean processes in general circulation models. .This class of models lacks an atmosphere and therefore external forcing data have to be specified. External forcing parameters typically include surface wind, surface downwelling radiation, temperature, humidity and precipitation. Data from the reanalysis projects such as NCEP/NCAR or ECMWF or operational forecasts are commonly used to force ice-ocean models. But the choice of forcing data is important: Small errors in the forcing data may translate into significant errors in model output and can easily mask natural variability. To date the NCEP/NCAR reanalysis has been the logical choice since it provided the longest time record. However, significant problems in the representation of the Arctic in the NCEP/NCAR reanalysis, particularly with respect to the radiation budget, have been noted. The recently completed ECMWF ERA-40 reanalysis now provides much hope for an improved forcing data set for ice-ocean model experiments. In this paper we examine the ERA-40 data set as a candidate atmospheric forcing data set in the context of our ongoing Arctic Sea-ice Ocean Reanalysis (ASOR) project. We compare forcing variables from the ERA-40 data set to NCEP/NCAR as well as in-situ and satellite validation data. We investigate the temporal and spatial variability of key variables and assess the impact of remaining uncertainties on ice ocean model results.