General Circulation Models (GCMs) typically exhibit systematic spatial and temporal errors in their representation of rainfall over southern Africa. The existence of systematic biases suggests the need to recalibrate GCM rainfall simulations. Model Output Statistics (MOS) is applied here to GCM simulated rainfall for the December-February season over southern Africa. This technique statistically recalibrates regional rainfall patterns by relating archived records from a number of GCMs’ (ECHAM4, NCEP-MRF9, COLA and NASA-NSIPP1) simulated rainfall fields to observations through a linear statistical technique. It has been shown before that a MOS approach improved on a single GCM’s deterministic (ensemble mean) rainfall forecasts over southern Africa. However, the inherent variability of the atmosphere requires seasonal climate forecasts to be expressed probabilistically. Furthermore, because models differ in their parameterizations, they differ in their performance under different conditions. Thus, simulated rainfall ensemble members of each GCM are combined into a multi-model MOS approach to produce probabilistic rainfall simulations for southern Africa’s mid-summer rainfall season. Using a suite of several models not only increases the effective ensemble size, it also leads to probabilistic forecasts that are skillful over a greater portion of the region and a greater portion of the time series. Finally, applying MOS to this super-ensemble removes remaining systematic biases, leading to the most reliable forecasts possible. Measures of multi-model MOS retro-active probability forecast skill are presented.