This paper discusses the behavior of the Arctic Ocean surface air temperature field for 1979–93. Temperatures are derived from a new gridded 6-h, 2-m air temperature dataset called POLES. These gridded air temperatures are estimated from optimal interpolation of temperature inputs from drifting buoys, manned Soviet North Pole (NP) drifting ice stations, coastal land weather stations, and ship reports. In processing the POLES data, the winter and summer properties of the mean NP temperatures are used to discard inaccurate or snow-covered buoys and to remove a summer warm bias. Comparison of the POLES and specific NP temperatures shows that the POLES temperature behaves well and gives better results than other gridded temperature datasets. Maps of the mean seasonal temperatures give realistic results consistent with other published estimates, and plots of the summer advance and retreat of the 0°C isotherm show the expected asymmetry between advance and retreat associated with the open water formati... Abstract This paper discusses the behavior of the Arctic Ocean surface air temperature field for 1979–93. Temperatures are derived from a new gridded 6-h, 2-m air temperature dataset called POLES. These gridded air temperatures are estimated from optimal interpolation of temperature inputs from drifting buoys, manned Soviet North Pole (NP) drifting ice stations, coastal land weather stations, and ship reports. In processing the POLES data, the winter and summer properties of the mean NP temperatures are used to discard inaccurate or snow-covered buoys and to remove a summer warm bias. Comparison of the POLES and specific NP temperatures shows that the POLES temperature behaves well and gives better results than other gridded temperature datasets. Maps of the mean seasonal temperatures give realistic results consistent with other published estimates, and plots of the summer advance and retreat of the 0°C isotherm show the expected asymmetry between advance and retreat associated with the open water formati...