The use of translog cost functions to represent the cost of production and/or the cost of energy has become quite commonplace in the last few years. Increasingly, the econometric results of such production/energy studies are being used for policy analysis. In light of this use, we address the question of the robustness of the translog representation during a period of rapidly changing relative energy prices. Since Canadian manufacturing data are available by region as well as nationally, there are several different methods of arranging the data for estimation: the data could be national, regional or pooled-regional. Using the Canadian data we explore the sensitivity of the translog representation of the production/energy model under national, regional and pooled data sets. The different methods of data arrangement yield significantly different parameter estimates.