Gross Primary Productivity (GPP) determines the initial substance and energy in the terrestrial ecosystem, which is an important link between the terrestrial carbon cycle and the atmospheric carbon pool. This study simulates the GPP in China from 1982 to 2004 by using the CLM4-CN (Community Land Model version 4 with Carbon-Nitrogen interactions) and evaluates its capability by comparing it with the MTE (Model Tree Ensemble)_GPP derived from upscaling of FLUXNET eddy covariance observations. We use the results to investigate the effects of different land cover datasets on GPP modeling. CLM4-CN is shown to effectively capture the spatial patterns of the GPP in China, which declines from southeast to northwest. However, the model overestimates the magnitude in most areas, particularly those south of 30°N. The annual GPP in China given by CLM4-CN (CLM4_GPP) is 13.7 PgC a^-1 on average; and that given by MTE_GPP is only 6.9 PgC a^-1. Although the CLM4_GPP and MTE_GPP show similar intra-annual cycles (R>0.9) for different dominant PFTs (Plant Functional Types) in China, the magnitude differs for most PFTs. Inter-annual variability in CLM4_GPP is higher than that in MTE_GPP for all PFTs. In addition, both the products show the same trends for tropical evergreen needleleaf trees, boreal deciduous broadleaf trees, and C3 grass; Differences are shown for the other PFTs. Precipitation is determined to be the main climate factor controlling temporal variation of the GPP in China during the experiment period. By modeling GPP using two different land cover datasets, we determine that different land cover datasets can cause obvious changes in the GPP for most regions in China. Thus, accuracy in the land cover dataset is important for the GPP simulation.