Four different wind products are evaluated in terms of their application to ENSO prediction. These wind products have been used to initialize an intermediate ocean-atmosphere coupled model for monthly retrospective forecasts from 1980 to 2002. The wind product that includes satellite scatterometer data has the highest scores, with the NCEP reanalysis and the new FSU objective analysis closely behind. The latter is a major improvement over the old FSU subjective analysis which has some serious problems in recent years. It seems that the wind products from remote sensing, in-situ observation and model reanalysis are all useful for ENSO prediction. At present, an ensemble of forecasts initialized with various wind data sets is probably our best bet.