This paper proposes an explanation of the shifts in the volatility of exchange rate returns that relies on standard present value exchange rate models. Agents are uncertain about the true data generating model and deal with the model uncertainty by making inference on the models and their parameters - a mechanism I call model learning. I show how model learning may lead agents to focus excessively on a subset of fundamental variables. As a result, exchange rate volatility is mainly determined by the dynamics of this subset of fundamentals. As agents switch between models the nominal exchange rate volatility varies accordingly even though the underlying fundamentals processes remain time-invariant. I investigate the relevance of this result empirically within the Taylor-rule based exchange rate model applied to the British Pound/US Dollar exchange rate. The results suggest that the observed change in volatility was triggered by a shift between models.