Most productivity indexes can be exhaustively decomposed into measures of technical change and efficiency change. Estimating these components usually involves the use of data envelopment analysis (DEA) or stochastic frontier analysis (SFA) models. This paper shows how assumptions concerning technologies, markets and firm behaviour can be used to frame these models. The paper explains that the assumptions underpinning common DEA models are rarely, if ever, true. On the other hand, the assumptions underpinning basic SFA models are almost always true. The parameters of basic SFA models can be estimated using ordinary least squares and two-stage least squares methods. More complex SFA models can be estimated using maximum likelihood methods. Unfortunately, the assumptions underpinning some of these more complex models are generally not true. This has important implications for estimating the drivers of productivity change. To illustrate, the paper uses common least squares and maximum likelihood methods to estimate the drivers of productivity change in U.S. agriculture. As expected, the different estimators lead to qualitatively different estimates of the efficiency change components productivity change.