Crop model is a widely-used strategy to evaluate the impact of climate change imposed on agriculture. The inter-comparison of gridded crop model is still at initial stage in China, thus making it crucial to comprehensively evaluate the performance of each global gridded crop model (GGCM). Using the statistics derived from Food and Agricultura Organization of the United Nations(FAO) and Ministry of Agriculture and Rural Affairs of the People's Republic of China(SYB) statistical rice yearly mean yield, this paper compared the simulated rice yields of 7 GGCMs (i.e. CGMS-WOFOST, CLM-CROP, EPIC-BOKU, GEPIC, LPJML, PDSSAT and PEPIC), which are driven by 2 climate datasets (AgMERRA and WFDEI-GPCC) and 3 management scenarios (Default, Fullharm and Harmnon) from 1980 to 2009 in China. The comparisons show that the simulated rice yields from different GGCMs have significant discrepancy in different regions of China. Each GGCM has different response and sensitiveness to climate datasets and management scenarios. The majority of simulations underestimate rice yield in China even though different statistical rice yield data will affect evaluation results. To some degree, GGCMs are able to simulate the inter-annual variation of rice yield and climate change effects, but hardly simulate the pattern of rice yield increase of statistics. The analyses of rice yield fluctuation on temporal and spatial aspect demonstrate LPJML and PDSSAT perform better among 7 GGCMs using 2 skill score approaches, and are most sensitive to different climate datasets and management scenarios, while CLM-CROP have lowest stimulation accuracy. In terms of management scenarios, the simulation on Default scenario performs dramatically better than that of Fullharm and Harmnon scenarios. In addition, Multi-gridded crop model ensemble (MME) could reduce simulation error compared to single GGCM, but requires suitable members to precisely perform MME.