A meta-analysis of animal feeding trials fed grass silages was conducted to study the effects of different forage harvests and maturity stages on feed intake and dairy performance in dairy cows. Data were collected from studies in Europe from 1981 to 2017. They were divided into 2 subsets: 69 diets from 11 studies were used for comparision of silages made from primary growth and regrowth of grass (harvest), and another 157 diets from 28 studies were used for comparision of digestibility influenced by the maturity of grass ensiled (D-value). The minimum prerequisite for an experiment to be included in the data set was that milk production (milk and ECM yields, concentration and yield of milk component), animal intake (DM, MP and ME), live weight (LW) and silage parameter (DM, CP, NDF, total acids, soluble N and D-value) were reported or could be estimated according to Finnish feed table. Additionally the harvest subset should involve a comparison of different cuts within the same experiment. Both subsets were analyzed using mixed model procedure of SAS. When comparing different harvests, the best fit model for milk protein concentration was: Milk protein concentration (g/kg) = 27.0 – 1.82 × a + (0.17 + 0.10 × a) × DMI (kg/d) + 0.004 × D-value (g/kg DM) (a = proportion of first cut silage in silage DM (0 or 1), Residual SE = 0.13, and AIC = 214). The best fit model for DMI was: DMI (kg/d) = – 11.3 + 1.43 × a – 0.005 × Silage soluble N (g/kg total N) + 0.057 × LW (kg) (a = proportion of first cut silage in silage DM (0 or 1), Residual SE = 0.16, and AIC = 186). When comparing the different maturity stages identified by D-value, the best fit model for ECM yield was: ECM yield (kg/d) = 16.0 + 0.105 × MEI (MJ/d) – 0.014 × D-value (g/kg DM) + 0.018 × silage CP (g/kg DM) – 0.031 × silage total acids (g/kg DM) – 0.007 × silage DM (g/kg DM) (RMSE = 0.84, R² = 0.95 and AIC = 462). The best fit model for DMI was: DMI (kg/d) = 8.81 + 0.03 × D-value (g/kg DM) – 0.01 × silage total acids (g/kg DM) (RMSE = 0.28, R² = 0.89 and AIC = 426). The prediction models from the meta-analysis could be used to improve the current ration formulation systems, or as prerequisites for economic models optimizing milk production in various farming systems.