Abstract: The product data based on GRACE gravity satellites provide important support for large-scale terrestrial water storage (TWS) research. However, these data cannot meet the needs of long-term sequence research because of the limited data length. Based on meteorological and hydrological observation data, a variable infiltration capability (VIC) model was constructed in the ten water resource zones in China. Based on the soil water and snow water storage output from the VIC model and meteorological observation data, an artificial neural network model based on a multilayer perceptron was developed to reconstruct a long-term (1980—2018), high-resolution (0.25° × 0.25°) monthly TWS anomaly (TWSA) dataset in China. The reconstructed TWSA data were evaluated using GRACE data from 2003 to 2018. The results demonstrate the following: ① The VIC model exhibits overall good simulation performance, with better performance in humid basins than in semi-arid basins.② The reconstructed TWSA dataset is highly consistent with the GRACE data at the spatial scale and can effectively capture interannual variations and evolution trends of the TWSA in most basins, especially in humid basins. ③ From 1980 to 2018, the TWSA exhibited a significant downward trend (>5 mm/a) in the North China Plain, Eastern Liaoning, Southern Songhua, and Southwest and Northwest parts of China, while a significant upward trend (>20 mm/a) was mainly concentrated in some regions of Western China. The TWSA data constructed can provide data support for hydrological and meteorological research in China.