The accuracy and precision of the National Research Council (NRC), Gesellschaft fur Ernahrungsphysiologie (GfE) and Institut National de la Recherche Agronomique (INRA) systems for predicting the digestible energy (DE) value of hays were determined from the results of 15 digestibility trials with natural grassland hays and 9 digestibility trials with lucerne hays that all met strict experimental and a tight corpus of methods. The hays were harvested in the temperate zone. They covered broad ranges of chemical composition and DE value. The INRA system was more accurate than the other two systems, with the bias between the predicted and measured DE values of natural grassland and lucerne hays averaging -0.11 and -0.04 MJ/kg DM with the INRA system, 0.34 and -0.70 MJ/kg DM with the NRC system and -0.50 and -1.69 MJ/kg DM with the GfE system (P< 0.05). However, the precision of the three systems was similar; the standard error of prediction corrected by bias was not significantly different (P> 0.05). The GfE system underestimated the DE value of hays, especially of lucerne hays. The differences between the predicted and measured DE values resulted mainly from the errors in the prediction of organic matter digestibility and energy digestibility for both natural grassland and lucerne hays. Discrimination according to botanical family (grasslandv. lucerne) can help improve the prediction of the DE value of hays. The choice of appropriate predictive variables is discussed in the light of differences in chemical composition and digestibility of the various cell wall components of grassland and lucerne hays. Neutral detergent fiber (NDF) may thus be preferable to ADF in the prediction equation of the DE value of lucerne hays, whereas ADF and NDF may both be relevant for natural grassland hays.