A number of hypotheses have been proposed to account for the role of lagged inflation in the New Keynesian price-adjustment model: (1) In the aftermath of abrupt structural change, rational learning may appear adaptive. (2) The model may have a serially correlated error term. (3) Estimating the model conditional on labor costs may remove or reduce the need for lagged inflation. I address the empirical support for these hypotheses and find that none eliminates the need for lagged inflation. In particular, lagged inflation enters with a coefficient in the range of 0.4 to 0.5, regardless of whether labor's share or detrended output is the measure of real marginal cost, or whether a serially correlated error term is allowed. Also, eliminating the period 1980-83 from the sample does not reduce the coefficient on lagged inflation.