This paper evaluates the exchange rate mechanisms in current large-scale models of the U.K. economy and finds that each has some shortcomings. A new, econometrically preferable, specification is developed. Previous empirical failures of exchange rate models relate to an inadequate treatment of expectations and neglect of the simultaneity between exchange rates and interest rates: the instrumental variable methods employed here remedy these deficiencies. The preferred equation, a forward-looking modified uncovered interest rate parity relation, outperforms a random walk. The sensitivity of overall model properties is examined by replacing the existing equations with the new equation and repeating standard simulation experiments.