In pasture-based dairy herds where silage is a widely adopted supplement, optimized feeding requires reliable estimations of nutritional quality of this conserved forage. Metabolizable energy, an important nutritional fraction, can be predicted from digestibility-related traits, such as the digestible organic matter contained in the dry matter (D-value). The aim of the present study was to evaluate the prediction of D-value and dry matter digestibility (DMD) of grass silages made from four different pastures and maturity stages, using the pepsin-cellulase method. Fungal cellulase was used, applying different enzyme concentrations, incubation times and types of final wash. The silages were prepared from permanent pasture (Dactylis glomerata L., Lolium perenne L., Bromus catharticus Vahl var. catharticus, Trifolium repens L. and Holcus lanatus L.), rotation pasture (Lolium multiflorum Lam. cv. Tama), oats (Avena sativa L.), and mixed pasture (L. perenne-T. repens). These were harvested at three different physiological stages (vegetative, ear emergence and dough grain). The treatment using an incubation time of 24 h, a cellulase concentration of 6.25 g L-1 and final wash with water (Treatment 3) presented the best prediction capacity of the in vivo D-value (R² = 0.78) and in vivo DMD (R² = 0.71). In vivo D-value prediction improved (R² = 0.8) when a chemical determination (crude fibre, gross energy, neutral detergent fibre, total ash or acid detergent fibre) was included in addition (multiple regression) to D-value obtained with cellulases (Treatment 3). Results of DMD obtained with cellulases show good precision, but underestimate in vivo values, and are closer to those obtained with ruminal fluid. Suitable equations could be used to improve accuracy.