Based on 36-year hindcasts from the fifth-generation seasonal forecast system of the European Centre for Medium-Range Weather Forecasts (SEAS5), the most predictable patterns of the wintertime 2-m air temperature (T2m) in the extratropical Northern Hemisphere are extracted via the maximum signal-to-noise (MSN) empirical orthogonal function (EOF) analysis, and their associated predictability sources are identified. The MSN EOF1 captures the warming trend that amplifies over the Arctic but misses the associated warm Arctic-cold continent pattern. The MSN EOF2 delineates a wave-like T2m pattern over the Pacific-North America region, which is rooted in the tropical forcing for the eastern Pacific-type El Nino- Southern Oscillation (ENSO). The MSN EOF3 shows a wave-like T2m pattern over the Pacific- North America region, which has an approximately 90° phase difference from that associated with MSN EOF2, and a loading center over mid-latitude Eurasia. Its sources of predictability include the central Pacific-type ENSO and Eurasian snow cover. The MSN EOF4 reflects T2m variability surrounding the Tibetan Plateau, which is plausibly linked to the remote forcing of the Arctic sea ice. The information on the leading predictable patterns and their sources of predictability are further used to develop a calibration scheme to improve the prediction skill of T2m. The calibrated prediction skill in terms of the anomaly correlation coefficient improves significantly over mid-latitude Eurasia in a leave-one-out cross-validation, implying a possible way to improve the wintertime T2m prediction in the SEAS5.