Aspects of seasonal forecast skill using global climate models (GCMs) are assessed over South Africa. The GCMs output is configured to predict low and high number of rainfall days exceeding predefined threshold values for the summer rainy seasons and to predict the rainfall totals of the onset of the rainy seasons for eight homogeneous rainfall regions of South Africa. Using canonical correlation analysis (CCA) as statistical downscaling tool the forecast skill levels of both coupled ocean-atmosphere and uncoupled atmospheric models are determined through retro-actively generated hindcasts. Both approaches have skill in predicting the low and high number of rainfall days exceeding predefined threshold values for the summer rainy seasons as well as rainfall totals of onset of the rainy seasons for the homogeneous rainfall regions. In addition to the forecast verification results, CCA pattern analysis is also performed to determine the dominating atmospheric circulation systems predicted to be controlling rainfall variations for the seasons of interest. CCA pattern analysis for both the GCMs indicate that when there are anomalously negative (positive) predicted 850hPa geopotential heights over South Africa, there are anomalously wet (dry) rainfall conditions over the most part of South Africa for the different seasons of interest. This work has paved the way for the operational production of seasonal rainfall characteristics over South Africa.