Reanalysis products from meteorological forecast centers perform well in areas where conventional data are dense, but the realism of their fields in data-sparse regions, such as the Arctic Ocean, is less certain. Upper-level winds from two reanalysis data sets, the NCEP/NCAR (National Centers for Environmental Prediction/National Center for Atmospheric Research) and the ECMWF (European Center for Medium Range Weather Forecasting) Reanalyses are evaluated in this study. Rawinsonde data from two Arctic field programs that were not assimilated by the models are compared to reanalysis wind products for five layers between 1000 and 300 hPa. Both reanalysis products exhibit large biases and are significantly too westerly and too northerly. Total wind speeds are too strong by 25 to 65% relative to rawinsonde values. Root-mean-square errors in u- and v-component winds are substantially larger than the mean values. Possible sources and ramifications of these discrepancies are discussed.