On the basis of monthly mean surface air temperature (SAT) data from 160 Chinese stations, the boreal winter is divided into early winter (ND, November and December) and late winter (JFM, January to March of the following year) by analyzing the relationship of SAT among different months. Following this definition, observational data from 160 Chinese stations, and reanalysis data from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) are studied by empirical orthogonal function (EOF) analysis, season-reliant EOF (SEOF) analysis, regression analysis and composite analysis to determine the interannual variations of early and late winter temperatures in China, their linkage, and related atmospheric circulations during the most recent 62 years. The first EOF modes in both early and late winter depict a pan-China temperature variation, and the second EOF modes in both early and late winter describe a temperature oscillation between northern and southern parts of China. An inspection of their corresponding principal components (PCs) indicates that if a certain SAT mode is observed in early winter, the possibility of observing the same SAT mode with the same sign in late winter is approximately 50%, which is nearly equal to that of observering the same SAT mode with the opposite sign. These results indicate that averaging the SAT for the entire winter is suitable for the former case and unsuitable for the latter case. Therefore, it is necessary to divide winter into early and later winters when the variability of wintertime SAT is investigated. To capture both the major modes of the SAT's interannual variability and the relationship between early and late winter, SEOF analysis is performed on the covariance matrix constructed with SATs from 160 Chinese stations recorded in both early and later winters. The first SEOF mode (SEOF1) captures an in-phase evolution of SAT from early to late winter. Its PC time series indicates a distinct interdecadal change in approximately the mid-1980s. The second SEOF mode (SEOF2) reflects an out-of-phase evolution of SAT from early to late winter. The corresponding PC time series is mainly featured with interannual variability. In the middle troposphere, the SEOF1 mode is closely associated with the Scandinavian teleconnection of the same sign throughout the winter, which consequently leads to changes in the intensity of the East Asian winter monsoon by altering the land-sea thermal contrast over East Asia and the intensity and the meridional location of the East Asian upper-tropospheric jet stream. In early winter, the SEOF2 mode is correlated to the Eurasian teleconnection, and the associated circulation anomalies are significant throughout the entire troposphere. In late winter, the Scandinavian-like teleconnection of the opposite phase dominates, and the significant circulation anomalies are observed only in the middle and lower troposphere. These results suggest that mid-latitude atmospheric internal dynamical processes, particularly atmospheric teleconnections, are the main causes of the formations of SEOF1 and SEOF2.