Surface air temperatures have recently increased more rapidly in the Arctic than elsewhere in the world, but large uncertainty remains in the time series and trend. Over the data-sparse sea ice zone, the retrospective assimilation of observations in numerical reanalyses has been thought to offer a possible, but challenging, avenue for adequately reproducing the historical time series. Focusing on the central Arctic Ocean, output is analyzed from 12 reanalyses with a specific consideration of two widely used products: the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), and the European Centre for Medium-Range Weather Forecasts interim reanalysis (ERA-Interim, hereafter ERA-I). Among the reanalyses considered, a trend of 0.9 K decade−1 is indicated but with an uncertainty of 6%, and a large spread in mean values. There is a partitioning among those reanalyses that use fractional sea ice cover and those that employ a threshold, which are colder in winter by an average of 2 K but agree more closely with in situ observations. For reanalyses using fractional sea ice cover, discrepancies in the ice fraction in autumn and winter explain most of the differences in air temperature values. A set of experiments using the MERRA-2 background model using MERRA-2 and ERA-I sea ice and sea surface temperature indicates significant effects of boundary condition differences on air temperatures, and a preferential warm bias inherent in the MERRA-2 model sea ice representation. Differences between experiments and reanalyses suggest the available observations apply a significant constraint on reanalysis mean temperatures.