Economic policy uncertainty (EPU) reflects more information related to exchange rate volatility than traditional macroeconomic variables do, and thus may have stronger predictive power in forecasting exchange rate volatility. Based on data from developed and developing economies (including Japan, the United Kingdom, the Euro Area, Brazil, the Russian Federation, India, and China), this paper provides in- and out-of-sample evidence that the EPU differential between two economies outperforms several macroeconomic predictors (for example, GDP growth and volatility of inflation growth) in forecasting exchange rate volatility, which is also supported by a Monte Carlo experiment. Moreover, the mechanism analysis shows that increased EPU differential is associated with increased foreign exchange trading activities and reduced bilateral trade, which helps explain the predictive power of EPU for exchange rate volatility. Furthermore, we find that EPU can predict both long- and short-run components of exchange rate volatility. Our main findings are robust after using alternative empirical proxies or controlling for endogeneity.