Three sources of uncertainty in model projections of precipitation change in China for the 21st century were separated and quantified: internal variability, inter-model variability, and scenario uncertainty. Simulations from models involved in the third phase and the fifth phase of the Coupled Model Intercomparison Project (CMIP3 and CMIP5) were compared to identify improvements in the robustness of projections from the latest generation of models. No significant differences were found between CMIP3 and CMIP5 in terms of future precipitation projections over China, with the two datasets both showing future increases. The uncertainty can be attributed firstly to internal variability, and then to both inter-model and internal variability. Quantification analysis revealed that the uncertainty in CMIP5 models has increased by about 10%–60% with respect to CMIP3, despite significant improvements in the latest generation of models. The increase is mainly due to the increase of internal variability in the initial decades, and then mainly due to the increase of inter-model variability thereafter, especially by the end of this century. The change in scenario uncertainty shows no major role, but makes a negative contribution to begin with, and then an increase later.