Abstract: This paper describes the key stochastic equations in the core model of the South African Reserve Bank (the Bank). During the development phase of the model, comments and suggestions from experts of other central banks, research institutions, and international and local academics were invited and incorporated, where feasible. As monetary policy influences the economy with time lags, it is important to have a view of future economic developments. Models are an indispensable tool in the monetary policy formulation process, providing a systematic framework for economic reasoning and helping to identify factors relevant in explaining mechanisms that could influence the economy in the future. A further benefit of models is that they allow the Bank to address uncertainty within a well-defined conceptual framework, and thereby examine the key risks associated with any forecast in a quantitative sense. The core model is used for forecasting purposes at Monetary Policy Committee (MPC) meetings and for simulation purposes, i.e. to quantify the impact of monetary policy decisions and possible shocks. The core model is a medium-sized Type II hybrid model, since it incorporates a long-run equilibrium that is based on economic theory and historical relationships, as well as short-run dynamics that allow the economy to gravitate towards its long-run equilibrium. The model uses official, seasonally adjusted quarterly data. In evaluating the model, overall model properties and performance receive preference over the properties of individual stochastic equations. An accuracy analysis of the Bank’s inflation forecasts and the monthly Reuters Survey, both covering the period 2000 to 2005, shows that CPIX inflation forecasts were mostly upwardly biased. However, the forecast errors produced by the Bank’s core model were consistently smaller and less biased than other institutions’ forecasts. Based on various peer reviews, the model appears to be a reasonable representation of the South African economy and is in line with international best practice for estimated structural models.