To investigate global and regional gross primary productivity (GPP) and its sources of uncertainties, widely used model structures of light use efficiency models are integrated to build a meta-model. Meteorological reanalysis data and remote sensing data are combined to estimate GPP, and a systematical and quantitative uncertainty analysis is conducted based on the ANOVA approach. Results show that: 1) the meta-model results correspond well with the upscaling of eddy-covariance measurements (FLUXCOM) GPP with a Pearson correlation coefficient of 0.97 and root mean square error of 24.36 gC/(m²·month) and outperforms any single combination of model structure. 2) Photosynthetically active radiation (PAR), water-related data and water regulation scalar (Ws) are the three main sources of uncertainties for global GPP estimates, contributing 41.73%, 26.79% and 23.82% respectively to total variance. 3) Sources of uncertainties of regional GPP depend on environmental conditions. For arid areas, Ws is the dominant contributor (over 80%). In cold areas, temperature regulation scalar (Ts) introduces over 40% of uncertainty. The findings not only highlight the necessity to reduce uncertainty of PAR and water-related data to reduce uncertainty in global and regional GPP estimates, but also point out the importance of improving performances of Ws and Ts algorithms under extreme environmental conditions.