Long-term, series gross primary production (GPP) data are important in carbon cycle research. The MOD17 algorithm, which is based on the radiation conversion efficiency concept of Monteith, has been used widely for estimating GPP. However, MODIS17 only provides the global GPP since 2000 due to the short time series of the MODIS leaf area index (MODIS15). LAI plays an important role in calculating the fraction of photosynthetically active radiation absorbed by plants, and errors in LAI will be propagated to GPP estimates. Three global LAI are available:MODIS15, improved-MODIS15, and Global Inventory Modeling and Mapping Studies (GIMMAS) LAI3g. The improved-MODIS15 LAI is more realistic and smoother than the MODIS15 because it uses quality control information and an integrated two-step method. The GIMMAS LAI3g is a new 30-year time series global LAI (1981-2011). In this study, we compared the global GPP estimates during 2000-2010 by using the MODIS GPP algorithm based on the three global LAI. The global GPP estimates based on GIMMAS LAI3g, MODIS15, and improved-MODIS15 are referred to as GPP_LAI3g, GPP_MOD15, and GPP_{impro_MOD15}, respectively. We also compared remote sensing-based GPP estimates with eddy covariance (EC) flux tower-measured GPP. The representative EC flux towers were selected by considering major typical plant functional types. We also analyzed spatio-temporal patterns and their correlations with the three GPP estimates as well as the MODIS17. The results showed the following. (1) The overall accuracy of the four global GPP estimates may be ranked as GPP_MOD17 > GPP_{impro_MOD15} > GPP_LAI3g > GPP_MOD15. (2) The four GPP estimates had high seasonal dynamic consistency. The estimated GPP values were closer to the flux tower-measured GPP in summer and winter than in spring and autumn. The accuracy of GPP_LAI3g was consistent for all seasons; GPP_MOD17 was more accurate than GPP_LAI3g for all seasons except for spring and fall. (3) GPP_LAI3g overestimated GPP for areas with moderate GPP values, i.e., the global total GPP value estimated by GPP_LAI3g was approximately (117±1.5) Pg C/a, which was higher than GPP_MOD17 and GPP_{impro_MOD15}. (4) The annual GPP values estimated by GPP_LAI3g were positively correlated with those by GPP_{impro_MOD15}, and approximately 63.29% of the global vegetated area had a significant correlation (P < 0.05). The GPP_LAI3g values were positively correlated with GPP_MOD15 in regions with low LAI uncertainty. Approximately 40.61% of the global vegetated area was significantly correlated with GPP_LAI3g and GPP_MOD15. There were also several negatively correlated areas, which may have been related to uncertainties and errors in the LAI and meteorological data. Based on our comparison, we conclude that GIMMS LAI3g is an effective dataset for GPP simulation at the global scale, and thus, the 30-year long-term GPP series estimated using the GIMMS LAI3g and MODIS GPP algorithms are reasonably acceptable.