The frequency of recurrence of drought has major societal, economical, and environmental impacts. However, our ability to capture drought conditions accurately are limited due to the uncertainties in current drought indices. In the present study, we proposed a Gravity Recovery and Climate Experiment (GRACE) total water storage (TWS) based drought severity index (DSI) using the detrended GRACE-TWS time series, to eliminate the effect of non-climatic factors on drought estimation and reflect true drought conditions. Based on the improved GRACE-DSI, we characterized the drought conditions over major basins in China during 2002–2017. Our results indicate that the improved GRACE-DSI can reasonably capture the drought process compared to existing non-detrended GRACE-based drought indices. The observed behavior of GRACE-DSI time series agrees reasonably well with the Palmer drought severity index, standardized precipitation index, and standardized runoff index, although differences exist due to intrinsic differences in the indicators of drought. Spatially, the Yellow River Basin, Huai River Basin, Hai River Basin, Southwest River Basin, and Continental River Basin share a similar pattern with droughts prevailing after 2013, and with both increases in duration and severity of the drought episodes. Moreover, pixel-based drought assessment also suggests an increasing trend in drought frequency in most basins in China during the GRACE era, with a prominent drought event in the Southwest River Basin beginning in April 2015 and ending in May 2016, with a severity of -25.38 and affecting 39.47 % of the total basin area. Our analyses demonstrate that the proposed GRACE-DSI can serve as a useful tool for integrated drought monitoring and provide a better understanding of drought conditions in major basins in China during 2002–2017.