This study investigated the linkage between winter temperature in the Yellow Sea (YS), China, and atmospheric indices and established this linkage through statistical models. The water temperature was obtained through hindcast simulation using a global–regional nested ocean model for the period of 1958–2007. The interannual variations of the simulated temperature were validated using satellite and in-situ observations. In the YS, the winter sea surface temperature (SST) had obvious interannual variations, with the maximum SST exceeding 2°C, and a significant shift from the cold to warm phase during 1988–1989. Based on the mechanism study, statistical models for the variations of water temperature in the YS were established using suitable atmospheric indices as predictors. For the northern YS (NYS) and the coastal region of the southern YS (SYS), statistical models of SST were established using linear regression based on the December–January–February mean Arctic oscillation index (AOI), representing the dominant large-scale atmospheric variability in boreal winter. For the YS warm current (YSWC) region, statistical models were established using both the AOI and the first principal component of the local wind stress curl (PC1-Curl), derived from the empirical orthogonal functions analysis. The PC1-Curl represents the influence of the local wind stress curl on the west-to-east shifts of the YSWC pathway. The applications proved that the models presented in this study have the ability to estimate winter temperatures in the YS within the recent years.