Difficulties in applying in vivo methods encourage the prediction of energy values of feeds in horses from alternative methods. Previous attempts to predict energy values from chemical composition have ignored differences among experiments. The current evaluation aimed to find the most reliable predictors (chemical composition and nutrient digestibilities) of apparent total tract energy digestibility of hays. Prediction was done with treatment means from a diverse collection of studies conducted worldwide. Experiment was included as a random effect in models to control for differences among experiments, and treatments means were weighted to account for unequal variances of the means across experiments. Organic matter digestibility (OMD; %) proved to be the most reliable (root mean square error [RMSE] = 1.6486; R2 = 0.8871) predictor of gross energy digestibility (GED; %) when calculated from 44 treatment means (n) obtained from 28 experiments conducted in 16 different studies (GED = 5.527 + 0.8124 OMD). Neutral detergent fiber (NDF; g/kg dry matter) presented a single chemical entity that was preferred over any multipredictor equation (GED = 70.4638−0.0360 NDF; RMSE = 2.9462; R2 = 0.6309; n = 58). However, the obtained equations should be validated against observed values, whereas alternative methods should be explored to predict OMD.