This paper addresses difficulties in modelling exchange rates in South Africa. Real exchange rate models of earlier research seem to be sensitive to the sample period considered, alternative variable definition, data frequency and estimation methods. Alternative exchange rate models proposed in this paper including the stock-flow approach and variants of the monetary model are not fully robust to data frequency and alternative estimation periods, either. Nevertheless, adding openness to the stock-flow approach and augmenting the monetary model with share prices and the country risk premium improves significantly the fit of the models around the large (nominal and real) depreciation episodes of 2002 and 2008. Interestingly, real commodity prices do not help explain the large depreciations. While these models do a reasonably good job in-sample, their out-of-sample forecasting properties remain poor.