Austral summer rainfall over the period 1991/1992 to 2010/2011 was dynamically downscaled by the weather research and forecasting (WRF) model at 9 km resolution for South Africa. Lateral boundary conditions for WRF were provided from the European Centre for medium-range weather (ECMWF) reanalysis (ERA) interim data. The model biases for the rainfall were evaluated over the South Africa as a whole and its nine provinces separately by employing three different convective parameterization schemes, namely the (1) Kain–Fritsch (KF), (2) Betts–Miller–Janjic (BMJ) and (3) Grell–Devenyi ensemble (GDE) schemes. All three schemes have generated positive rainfall biases over South Africa, with the KF scheme producing the largest biases and mean absolute errors. Only the BMJ scheme could reproduce the intensity of rainfall anomalies, and also exhibited the highest correlation with observed interannual summer rainfall variability. In the KF scheme, a significantly high amount of moisture was transported from the tropics into South Africa. The vertical thermodynamic profiles show that the KF scheme has caused low level moisture convergence, due to the highly unstable atmosphere, and hence contributed to the widespread positive biases of rainfall. The negative bias in moisture, along with a stable atmosphere and negative biases of vertical velocity simulated by the GDE scheme resulted in negative rainfall biases, especially over the Limpopo Province. In terms of rain rate, the KF scheme generated the lowest number of low rain rates and the maximum number of moderate to high rain rates associated with more convective unstable environment. KF and GDE schemes overestimated the convective rain and underestimated the stratiform rain. However, the simulated convective and stratiform rain with BMJ scheme is in more agreement with the observations. This study also documents the performance of regional model in downscaling the large scale climate mode such as El Nino Southern Oscillation (ENSO) and subtropical dipole modes. The correlations between the simulated area averaged rainfalls over South Africa and Nino3.4 index were −0.66, −0.69 and −0.49 with KF, BMJ and GDE scheme respectively as compared to the observed correlation of −0.57. The model could reproduce the observed ENSO-South Africa rainfall relationship and could successfully simulate three wet (dry) years that are associated with La Nina (El Nino) and the BMJ scheme is closest to the observed variability. Also, the model showed good skill in simulating the excess rainfall over South Africa that is associated with positive subtropical Indian Ocean Dipole for the DJF season 2005/2006.