Thermoelectric power generation accounts for over 41% of total U.S. freshwater withdrawals, making understanding the determinants of power plants’ water withdrawals (WW) and consumption (WC) critical for reducing the sector’s reliance on increasingly scarce water resources. However, reported data inconsistencies and incomplete analysis of potential determinants of thermoelectric water use hinder such understanding. We address these challenges by introducing a novel data filtering method and a more complete assessment of water use determinants. First, we applied a power-cooling ratio as an operations-based data filter that removed operationally implausible records while retaining more original data, outperforming previous statistical filtering methods. Second, we found that different water use reporting methods (WURMs) provided statistically significantly different WW and WC values, revealing the importance of this previously unrecognized feature in reported water use records. Third, our data-driven approach showed that traditionally emphasized features─such as cooling technology and gross generation─are of primary importance but can be surpassed by other, often overlooked, features when modeling WW or WC individually. The plant configuration, cooling technology, and gross generation were the most important features of WW, whereas WURM, cooling technology, and reporting month were the most important for WC. These findings can improve thermoelectric power plant management, water use reporting accuracy, and water use modeling.