With a three-dimensional variational (3D-Var) assimilation scheme and a hybrid coupled model, we have explored the possibility of initializing ENSO prediction models by assimilating NCEP (National Centers of Environment Prediction) reanalysis subsurface temperatures. Our results show that, compared to predictions without assimilation, the reanalysis product can effectively improve prediction of both Ni¿o3 sea surface temperature anomalies (SSTA) at all lead times up to 12 months (in particular for lead times over 4--6 months) and of El Ni¿o episodes. The oceanic analysis from the assimilation with the reanalysis product can be as good as those generated by directly assimilating subsurface in situ temperature observations.