A strong lagged relationship between El Niño–Southern Oscillation (ENSO) and rainfall in the main rain season (October–January) on the leeward islands of Aruba, Curaçao and Bonaire is found. It can easily be used for skilful seasonal predictions, with an anomaly correlation coefficient r ≈ 0.6 at lag 4 months on historical data. The other two seasons, February–May and June–September, also show correlations with ENSO that can be exploited for predictions, r = 0.4 to 0.5. In the February–May dry season there is also a lagged correlation with sea-surface temperature (SST) in the Pacific Ocean off the Central American coast that can be used to increase the forecast skill. A June–September small rains season correlation to equatorial Atlantic Ocean SST is absent in earlier data. Most of these results are also applicable to other stations in northern South America. Regressions with the circulation show that the main intermediate factors are upper-level divergence and vorticity, and at lower levels a veering of the trade winds. This modifies the descending limb of the sea–continent breeze circulation that is responsible for the dry zone off the coast. Copyright © 2002 Royal Meteorological Society.