Based on generalized extreme value distribution (GEV),the statistical model of precipitation extremes was established,the extreme precipitation events was defined by the recurrence period of 20 years and the bias correction and spatial disaggregation (BCSD) downscaled CMIP5 data were employed to fit the GEV distribution,aiming to quantify and evaluate the temporal and spatial variations of precipitation extremes of the nine major basins of China in 21st century,and to assess the simulation performance on precipitation extremes of downscaled CMIP5 data over China. The correlation between extreme and average precipitations was also discussed. The results show that,the CMIP5 GCM models simulated precipitation extremes over China reasonably well,which ensured a basically high level of credibility. The historical extreme monthly precipitation manifested a distribution pattern of more in the southeast and less in the northwestfor the period of 1901 - 2005,and the future extreme monthly precipitation was projected to increase by 3.2% - 12.8% and 7.7% - 24.7% in the first and second half of the 21st century,respectively. The increasing variation pattern was quite similar to the spatial distribution pattern of high - low - highfrom Northwest to Southeast. The increasing radiation forcing concentration would trigger upward variations in precipitation extreme magnitudes,especially in the second half of 21st Century. As for the correlation between the changes of extreme and average precipitations,the northern and central basins exhibited positive correlation,whereas the southern basins of Southeastern Rivers Basin and Pearl River Basin indicated partial negative correlations.