The predicability of the El Ni¿o/Southern Oscillation (ENSO) phenomenon was investigated by analyzing observed sea levels, surface stresses, and subsurface temperatures simulated with an oceanic general circulation model forced by observed winds. In addition, a large ensemble of prediction experiments has been conducted with a simplified coupled ocean-atmosphere model consisting of an oceanic general circulation model coupled to a simple atmospheric feedback model. Our analysis supports the hypothesis that the ENSO-related interannual variability in the tropical Pacific can be understood as a cycle within the coupled ocean-atmosphere system which is inherently predictable. This cycle consists of an accumulation of warm water in the western Pacific during the cold phases of ENSO and a loss of this heat during its warm phases. The results of the prediction experiments with our simplified coupled ocean-atmosphere model indicate that the phase of tropical Pacific sea surface temperatures is predictable two to three seasons in advance with our simplified coupled system, whose dynamics is governed by the ocean. We found a strong dependence of the skills on season, with spring SSTs being least predictable.