Testing for purchasing power parity (PPP) and uncovered interest parity (UIP) has been the focus of many empirically oriented studies. Whilst these simple economic theories of exchange rate and interest rate determination are theoretically attractive, the empirical support for these equilibrium conditions is at best mixed. Many potential reasons have been cited in the literature for the failure of such studies, ranging from market imperfections to inappropriate modelling strategies. The current state-of-the-art procedure involves testing for two cointegrating vectors in a multivariate error correction model which may be economically identified as the PPP and UIP relations. However, such a procedure does not account for the policy regime shifts which often characterise economic time series. It is proposed that such regime shifts distort the underlying PPP and UIP relations, making them difficult to detect when modelled within this conventional framework. In this paper, a Markov-switching vector error correction model (VECM) is considered for time series data in which monetary and exchange rate regime shifts are known a priori to be present. Weak evidence in favour of PPP and UIP is established in a standard linear VECM, although the residuals of this model clearly indicate that it is inappropriate in terms of functional form. The Markov-switching VECM, however, provides convincing evidence in favour of both the PPP and UIP relations and a marked improvement in the residual distributions. An enlightening by-product of this method is the databased estimation of regimes which appear to conform to those suggested by economic history.