The superensemble (SUP) hindcastof the surface air temperature over East Asia and its surrounding areas was conducted based on the CMIP5 runs of 15 climate system models. The ERA data of the monthly surface air temperature were used as observed values, and root-mean-square error (RMSE), anomaly correlation coefficient (ACC), and mean absolute error (MAE) were chosen to evaluate the hindcast skills of individual models, the multimodel ensemble mean (EMN) and SUP techniques. The results show that the length of the training period has influence on the hindcast results. The RMSE decreases and the ACC increases as the training period length increases within some ranges. However, the hindcast errors will not decrease or even increase when the training period reaches a particular length. All of the 15 individual models can hindcast with some skills the surface air temperature over East Asia and its surrounding areas, its interannual variation and spatial distribution. Nevertheless, the hindcast skills are different for different models, among which the CCSM4 model has the highest hindcast skill. Multimodel ensemble techniques are superior to individual models in terms of the hindcast skills. The SUP technique has higher hindcast skill than the EMN. The regionally averaged RMSE of the SUP hindcast is around 0.43℃, suggesting that the SUP improves the hindcast skill of the surface air temperature considerably.