A combination of genetic algorithm (GA) and empirical orthogonal function (EOF) analysis has been employed to forecast satellite scatterometer–observed surface winds in the north Indian Ocean. The EOFs are used to compress the major part of the spatial variability into a few eigenmodes. The temporal variability is contained in the corresponding principal components (PC) which have been subjected to singular spectrum analysis for filtering out the random part. Forecast of the deterministic part has been carried out using GA with lead times varying from 1 to 4 days. The entire analysis has been done separately for the zonal and meridional winds. Finally, predicted wind fields have been generated as linear combinations of the spatial eigenmodes weighted by the predicted PCs. Forecast quality has been evaluated by comparing with an independent validation data set as well as with buoy data. The performance of the algorithm has been found to be quite encouraging.