With a simple 3D Var assimilation scheme, a new strategy for assimilating sea surface temperature (SST) observations has been proposed in this paper. The strategy involves assimilating two proxy data, SST and subsurface thermal data, into the ocean model. An ensemble of predictions for the Ni¿o3 region SST anomalies (SSTA) is performed to validate the new strategy. The results show that the new strategy can effectively improve Ni¿o3 SSTA predictions at all lead times, in particular for lead times over 6 months, and for the predictions of El Ni¿o episodes. The prediction skills of the Ni¿o3 SSTA attained by the new scheme can be as high as those attained by the assimilation of subsurface data and sea level height. Comparisons between two schemes of SST assimilations suggest that the impact of observations on the initializations of ENSO predictions could greatly depend on how the observations were assimilated.