Abstract: The evaluation of the applicability of precipitation indices is a critical component in flood impact analysis. Focusing on annual flood loss across 31 provinces of China from 2002 to 2023, this paper constructs both stationary and nonstationary loss models and comparatively evaluate the applicability of 13 commonly used precipitation indices using goodness-of-fit (R²), Bayesian information criterion (BIC), and root mean square error (E_RMS). The results show that:①The 12-month standardized precipitation index (SPI-12) exhibits overall the highest applicability across the full sample, yet its performance is relatively unstable in high-loss years. Under the nonstationary model, the median R² values for affected population and direct economic loss are 0.50 and 0.52, respectively. The simple daily intensity index and the number of days with daily precipitation ≥ 20.0 mm show slightly lower applicability than SPI-12, as they fail to account for the distribution of multi-year precipitation—their median R² values for affected population are 0.48 and 0.46, and for direct economic loss are 0.36 and 0.38, respectively. ②Despite the inclusion of a time term in the nonstationary model, the median BIC values of the precipitation indices remain comparable to or lower than those of the stationary model. ③Compared to area weighting, the exposure-based weighting (population and GDP) helps to reduce the E_RMS of some precipitation indices in fitting flood loss, especially annual total wet-day precipitation and precipitation anomaly. Overall, the rational selection of precipitation indices is important for flood impact assessment.