There are two main approaches to modelling monetary policy; simple instrument rules and optimal policy. We propose an alternative that combines the two by extending the loss function with a term penalizing deviations from a simple rule. We analyze the properties of the modified loss function by considering three different models for the US economy. The choice of the weight on the simple rule determines the trade-off between optimality and robustness. We show that placing some weight on a simple Taylor-type rule in the loss function, one can prevent disastrous outcomes if the model is not a correct representation of the underlying economy.