Abstract: The remotely sensed burned area (BA) products can provide continuous and spatiotemporally explicit characteristics of fire patches, which are critical data sources for understanding regional fire regimes. However, their accuracy remains to be improved. In this study, a global BA product (i.e., CCI_Fire) at 250 m resolution was integrated with global forest change (GFC) product at 30 m to generate a refined BA product, named CCI_GFC product, whose accuracy was evaluated through comparing the BA with pre-existing fire patches data. To reveal the characteristics of forest fire in China between 2001 and 2017, we conducted a grid analysis at 0.05°×0.05° spatial resolution based on the refined BA product and the spatial pattern of eco-regions at the macro scale. The results showed that the accuracy metrics including the recognition rate (RR), variance explained (R²), root mean squared error (RMSE), mean absolute percentage error (MAPE) of the CCI_GFC product (i.e., 83%, 0.91, 0.28, and 8.5% respectively) were all superior to the CCI_Fire product (i.e., 74%, 0.86, 0.36, and 11.8% respectively) and the MCD64A1 product (i.e., 35%, 0.78, 0.48, and 17.3% respectively). The total burned area of forest was approximately 12.11 million hm² for the whole country from 2001 to 2017, while the annual burned area temporally decreased. Forest fires in China were dominated by the low-frequency [0<burned forests rate (BFR)≤40%] burns which occupied 79% of the total burned area. The medium-frequency (40%<BFR≤80%) and high-frequency (80%<BFR≤120%) burns accounted for 11% and 10% of the total burned area, respectively. The seasonality of forest fires differed substantially among eco-regions. The primary fire season of the temperate eco-region was spring (May) and autumn (October). Fire season in the tropical-subtropical eco-region started from the winter to the next spring which included the months from November to next March.