Prediction skills for southern African (16˚-33˚E, 22˚-35˚S) summer precipitation in the Scale Interaction Experiment-Frontier coupled model is assessed for the period of 1982-2008. Using three different observation datasets, deterministic forecasts are evaluated by anomaly correlation coefficients, whereas scores of relative operating characteristic and relative operating level are used to evaluate probabilistic forecasts. It is shown that these scores for forecasts of December-February precipitation initialized on October 1 st are significant at 95%confidence level. On a local scale, the prediction skills in the northwestern and central parts of southern Africa are higher than those in northeastern South Africa. El Niño/Southern Oscillation (ENSO) provides the major source of predictability, but the relationship with ENSO is over-confident in the model. Also, the Benguela Niño, the basin mode in the tropical Indian Ocean, the subtropical dipole modes in the South Atlantic and the southern IndianOceans and ENSO Modoki may provide additional sources of predictability. When prediction skills are evaluated for the whole wet season from October to the following April, it is found that precipitation anomalies in December-February are most predictable. The present study presents promising results for seasonal prediction of precipitation anomalies in the extratropics, where seasonal forecast are considered a difficult task.3