Much of the existing literature on the relationship between nuclear energy consumption and gross domestic product (GDP) deals only with the causal links between these two variables. However, very little attention has been paid to the structure and form of this relationship. This paper first uses panel cointegration techniques to illustrate the form of an inverted U-shaped curve that arises from pooled data, then, applies the iterative empirical Bayesian procedure in order to account for the heterogeneity in the coefficients of the long-term relationship. The empirical results from a multivariate framework involving carbon dioxide (CO2) emissions reveal that for only 3 of the 21 nuclear countries studied, a linear form of the relationship can be justified and that nuclear energy goes from being a normal good to being an inferior good for the majority of the sample countries. (C) 2013 Elsevier Ltd. All rights reserved.