Reservoir is essential for water resources management and utilization, and its storage capacity is the key information for the scientific and engineering community. Over 95 % of the ～ 100,000 reservoirs constructed in China lack accurate capacity information due to the high cost and difficult accessibility of field measurements. Here we develop models of reservoir capacity estimation based on statistical and machine learning (ML) algorithms for the national-scale reservoirs in China. Based on the collected capacity records of a small portion of reservoirs and surrounding environment factors, we built statistical models by considering their associated variables, as well as ML models with different algorithms to predict the unknown reservoir capacities. The prediction results from different statistical and ML models were evaluated by comparing with the validated samples, and the optimal model was selected to estimate the reservoir capacity for further analyses. Results show that among the four statistical models and six ML models, the ML model using the extreme-gradient-boosting-tree (XGBoost) algorithm performs best in capacity prediction with a correlation coefficient of 0.97, a Nash-Sutcliffe efficiency of 0.95, and a mean absolute percentage error of 22.93 %. The estimated capacity of unrecorded reservoirs using XGBoost was 279.28 km3 (254.52–306.47 km3), and the total capacity of Chinese reservoirs was 1,065.19 km3 (1,040.43–1,092.38 km3). The proposed method provides a feasible approach of rapid estimation of reservoir capacity and monitoring of storage variations at large scales, which is critical for the rational conservation and utilization of water resources and the formulation of related policies.