In this paper we adopt a Bayesian stochastic frontier framework for measuring the components of output growth in a set of 17 OECD countries. Empirical results indicate that all three components play an important role in explaining output growth. However, it was difficult to find a general pattern which could form the basis for universal policy conclusions with respect to productivity growth. We note that our work is similar in spirit to the growth accounting literature. Although too voluminous to cite here (see Maddison (1987) for a survey), this literature tends to lump the unexplained residual under the rubric 'technical change'. Using a stochastic frontier model, however, enables us to analyze efficiency issues formally and it allows us to give a structural interpretation to our unexplained residual. Thus, we address the issue of output growth in a somewhat more fundamental way than in the growth accounting literature. Furthermore, our work also relates to the enormous literature that seeks to explain economic growth using cross-country regressions (e.g. Barro (1991), DeLong and Summers (1991), Levine and Renelt (1992), Persson and Tabellin (1994)). Unlike this literature, however, we focus on measurement rather than explanation, using a simple economic model and then seeing what insights can be derived through the use of careful statistical methods. This is in contrast to cross-sectional regression approaches that seek to consider a myriad of possible deeper 'structural' reasons for empirical findings. In statistical terms, we are interested in investigating the properties of the distribution of output conditional on capital and labor. Researchers who perform cross-country growth regressions implicitly argue that the distribution of output conditional on capital, labor, and many other complex variables, is the more appropriate focus for studies into productivity growth. However, the investigation of such a distribution typically involves selecting out only a few of the potentially enormous number of conditioning variables. For computational ease researchers engaged in cross-country growth studies will usually assume a linear relationship along with minimal dynamics and a simple error structure. Given the restrictiveness of such a statistical model and the lack of robustness in cross-country growth regressions (Levine and Renelt (1992)), we would argue that our approach is a sensible complement. In addition, we start from models with a theoretical foundation in production theory and imposing the economic regularity conditions is an important step in avoiding ad hoc methods, based exclusively on data-mining. Allowing for model and parameter uncertainty is of enormous importance in any empirical study since it enables the researcher to guard against drawing strong conclusions from weak evidence. In this paper, we have made many conclusions and recommendations based on our empirical results but have stressed that both model and parameter uncertainty must lead us to qualify significantly the recommendations we make. That is, substantial standard deviations for quantities of interest and a certain degree of sensitivity to model choice means that our conclusions should be taken tentatively and within the full understanding of these limitations. Note, however, that these limitations are not specific to our particular approach since any sensible study would doubtlessly reveal a similar sensitivity. The limitations of the present data allow us to draw only tentative conclusions about the macroeconomic debates we consider, and to pretend otherwise would surely be misleading. [ABSTRACT FROM AUTHOR]