Accurately predicting the net energy value of compound feeds fed to ruminants is a major challenge as the precise composition of ingredients is usually unknown. The data set "CompEnerRum" compiled in vivo data of 83 compound feeds obtained in one of four European research units (two in France, one in The Netherlands and one in the United Kingdom) associated with several laboratory analyses. The accurate ingredients composition of these compound feeds was known. This dataset exhibited a broad spectrum of ingredients with a wide range of energy digestibility which covered almost all the feeding situations found in European farms. The in vivo organic matter and energy digestibility, and for two units, energy losses as methane or urine data of compound feeds were obtained either by difference (when the roughage digestibility was also measured) or by regression (measurements with several incorporation percentages of the compound feed). Gross energy, ash, crude protein, crude fibre, ether extract, cell wall (NDF and ADF) and lignin (ADL and Christian) contents were determined. Two enzymatic methods were also performed. All these data might be used to estimate net energy value of feeds, as this stepwise method of energy losses is the basement of all the energy systems in ruminants.