Digestible energy (DE) accounts for the major fraction of the energy content of hay fed to equines, but DE measurements are difficult to carry out, and researchers have developed models for predicting the DE content of hays for horses from their chemical composition. Digestible energy varies depending on botanical family, species, and agronomic and climatic conditions. A database of 61 hays harvested in a temperate zone (32 natural grassland hays, 16 grass hays and 13 lucerne hays) was set up from studies carried out at the French National Institute of Agricultural Research (INRAe) or extracted from the literature using a similar methodological corpus, to test the relevance of crude fibre (CF), NDF, ADF, total free reducing sugars (TFRS), CP as potential criteria for the prediction of organic matter digestibility (OMD), energy digestibility (ED) and digestible energy (DE). Models were developed for natural grassland hay, grass hay and lucerne hay. The adjusted cross-validation R2 values (R2CV) of the models obtained using CF, ADF or NDF as single variables or in combination with CP for the prediction of OMD, ED and DE of natural grassland hay ranged from 0.84 to 0.86, from 0.83 to 0.86 and from 0.82 to 0.84, respectively. For grass hay, the R2CV were lower and ranged from 0.65 to 0.71, from 0.63 to 0.71 and from 0.62 to 0.68, for the prediction of OMD, ED and DE, respectively. Finally, for lucerne hay, the R2CV ranged from 0.65 to 0.72, from 0.34 to 0.74 and from 0.25 to 0.67, for the prediction of OMD, ED and DE, respectively. The R2CV of the prediction models of ED and DE were more heterogeneous for lucerne hay than for grassland and grass hay. The TFRS content of hay was never a significant predictive variable of OMD, ED or DE, either as single variable or combined with any cell wall content criterion. The models obtained in the present study improved those proposed by INRA to date.