As the real exchange rate of developing countries is especially vulnerable to stochastic events, standard unit root tests do not capture such events adequately. Using a Markov switching extension of the ADF test, which incorporates stochastic regime switching, we address the issue of real exchange rate stationarity for 43 developing countries. We find strong statistical evidence that this approach is preferred to the standard ADF for all countries considered. For 36 countries, there is strong evidence of regime-dependent stationarity, namely there is a regime in which the real exchange rate is stationary and another regime in which the real exchange rate is non-stationary. This suggests that over a sample period, there are sub-periods of stationarity and sub-periods of non-stationarity. We identify those sub-periods and assess their average duration and regime persistence. The results, robust to alternative sample periods, indicate that there exists sample-dependence in unit root results in previous studies, and help bridge the gap between conflicting results of these studies. Copyright © 2008 John Wiley & Sons, Ltd.