The present article offers a careful description of empirical identification of possible multiple changes in regime. We apply recently developed tools designed to select among regime-switching models among a broad class of linear and nonlinear regression models and provide a discussion of the impact on the formation of inflation expectations in the presence of multiple and recurrent changes in inflation regimes. Our empirical findings give a plausible explanation as to why the rational-expectations hypothesis based on direct measures of inflation expectations from survey series is typically rejected because of large systematic differences between actual and expected inflation rates. In particular, our results indicate that in the case of changing and not perfectly observed inflation regimes, inference about rationality and unbiasedness based on a comparison of ex ante forecasts from survey series and actual inflation rate based on ex post realizations will be ambiguous because of the presence of an ex post bias. The empirical findings are based on Danish inflation rates covering 1957-1998. We show that it is not possible to reject the hypothesis of multiple inflationary regimes and that the actual inflation rate can be represented by a two-state Markov regime-switching model. It turns out that the real-time forecasts produced from this model exhibit a large degree of similarity when compared to the direct measures of inflation expectations. The result illustrates the important impact of switching regimes on the formation of actual and expected inflation and hence of ex post bias as a main contributor to the difference between actual and expected inflation observed directly from survey series.