We review the changes in modelling strategy and econometric methodology when estimating a firin-level investment equation on panel data during the past twenty years, in order to assess which of these changes result from new estimation methods and changes in the practice of panel data econometrics, and which are "real" and due to the evolution of the economy. Thus our paper consists of a series of comparisons: a simple accelerator-profit specification versus one with error correction, traditional between- and within-firm estimation versus GMM estimation, the investment behavior of Rench firms versus that of U.S. firms, and investment behavior today versus ten to twenty years ago. Although the econometric advances have perhaps not been as successful as we had hoped, we do find some real change in firm behavior and some improvement in equation specification and interpretation during the past twenty years.