Abstract The Indian Ocean Dipole (IOD) has profound socioeconomic impacts on not only the countries surrounding the Indian Ocean but also various parts of the world. A forecast system is developed based on a relatively high-resolution coupled ocean–atmosphere GCM with only sea surface temperature (SST) information assimilated. Retrospective ensemble forecasts of the IOD index for the past two decades show skillful scores with up to a 3–4-month lead and a winter prediction barrier associated with its intrinsic strong seasonal phase locking. Prediction skills of the SST anomalies in both the eastern and western Indian Ocean are higher than those of the IOD index; this is because of the influences of ENSO, which is highly predictable. The model predicts the extreme positive IOD event in 1994 at a 2–3-season lead. The strong 1997 cold signal in the eastern pole, however, is not well predicted owing to errors in model initial subsurface conditions. The real-time forecast system with more ensembles successfully predicted the weak negative IOD event in the 2005 boreal fall and La Niña condition in the 2005/06 winter. Recent experimental real-time forecasts showed that a positive IOD event would appear in the 2006 summer and fall accompanied by a possible weak El Niño condition in the equatorial Pacific.