This paper analyzes energy use and CO2 emissions of more than 78 000 German industrial establishments between 1995 and 2006. It is the first study to exploit exceptionally rich energy data that were recently matched to official micro datasets. We document that both energy use and intensity are highly dispersed across establishments. When isolating the between-sector variation in energy intensity, there is a strong positive correlation with energy use, CO2 emissions and emissions intensity. Yet there is no evidence that the scale of an industry determines its energy intensity. The dispersion of energy use across establishments of a given sector, normalized by the median, is positively correlated with that of gross output, but not with the median energy use. Similarly, there is no evidence that the median energy intensity is correlated with the within-sector dispersion of energy intensity or with that of CO2 emissions. Looking at the fuel mix across sectors, we find that more energy intensive industries rely more on fuels other than electricity, although the variability among establishments in those industries is extremely high. We also demonstrate that average fuel shares are sensitive to the skewness of the underlying distribution and recommend the use of median fuel shares for better representativeness.