We use meteorological measurements from three drifting buoys to evaluate the performance of the ERA-Interim and ERA5 atmospheric reanalyses from the European Center for Medium-Range Weather Forecasts over the Weddell Sea ice zone. The temporal variability in surface pressure and near-surface air temperature is captured well by the two reanalyses but both reanalyses exhibit a warm bias relative to the buoy measurements. This bias is small at temperatures close to 0°C but reaches 5–10°C at −40°C. For two of the buoys the mean temperature bias in ERA5 is significantly smaller than that in ERA-Interim while for the third buoy the biases in the two products are comparable. 10 m wind speed biases in both reanalyses are small and may largely result from measurement errors associated with icing of the buoy anemometers. The biases in downwelling shortwave and longwave radiation are significant in both reanalyses but we caution that the pattern of bias is consistent with potential errors in the buoy measurements, caused by accumulation of snow and ice on the radiometers. Overall, our study suggests that, with the exception of near-surface temperature, both reanalyses reproduce the buoy measurements to within the limits of measurement uncertainty. We suggest that the significant biases in near-surface air temperature may result from the simplified representation of sea ice used in the reanalysis models, and we recommend the use of a more sophisticated representation of sea ice, including variable ice and snow thicknesses, in future reanalyses.