Abstract: This study presents a comprehensive evaluation of three modern precipitation datasets—the NOAA gauge-corrected Multi-Radar Multi-Sensor (MRMS) product, the final run of NASA’s Integrated Multi-satellite Retrievals for GPM (IMERG-Final), and the NCAR–USGS convection-permitting hydroclimate reanalysis CONUS404—based on 25 extreme rainfall events over the contiguous United States (CONUS) during 2015–2024. These storm events were selected from NOAA’s storm events database such that, for at least one accumulation period, the observed precipitation over a large area has an annual exceedance probability (AEP) of 1/500 or less. Using NOAA’s AORC, a high-resolution gridded near-surface forcing dataset for CONUS-wide hydrologic modeling, as the benchmark, we evaluate the performance of the three datasets in terms of storm-total accumulations, temporal variations of area-mean rainfall, and hourly rain rates at both regional and pixel scales. Evaluation metrics include the Pearson correlation coefficient (PCC), relative bias (RB), and root-mean-square error (RMSE) for rainfall magnitude, and probability of detection (POD), false-alarm ratio (FAR), and Heidke Skill Score (HSS) for rainfall detection. Results show that MRMS performs best overall among the three datasets. For both accumulated rainfall and hourly rain rates, its PCC values consistently exceed 0.9, indicating that MRMS can reproduce the spatial and temporal patterns of extreme rainfall events with high accuracy. Its hourly rainfall detection capability is also strong, with HSS generally above 0.8 across different storm events. These characteristics highlight the suitability of MRMS for high-resolution monitoring and research on extreme rainfall. IMERG-Final demonstrates reasonable hourly rainfall estimation, with PCC values around 0.5 relative to the AORC reference, and moderate detection skill, with HSS between 0.6 and 0.8 for most events. It can capture the general spatial and temporal patterns of many extreme rainfall events. Nevertheless, IMERG-Final exhibits noticeable magnitude-dependent biases: heavy rainfall is significantly underestimated. For example, for the 95th percentile of storm totals, IMERG-Final tends to underestimate amounts by 79.6% across all events. For hourly rainfall, it also tends to underestimate pixel-scale rain rates overall, with an RB of −12.3%. These biases indicate that IMERG-Final should be used with caution for extreme rainfall estimation and design-storm analysis, unless additional bias-correction is applied. In contrast, the convection-permitting reanalysis CONUS404 demonstrates superior performance in estimating high-percentile accumulated rainfall at the event scale. For the 95th percentile of storm-total accumulation, its PCC against AORC can reach approximately 0.96, and RB is around −40%. Its estimation of hourly rain rates shows a relatively small negative bias (≈−1.4%), suggesting that CONUS404 precipitation intensity is not strongly biased on average. However, spatial displacement errors in the simulated precipitation fields lead to pronounced discrepancies in pixel-scale comparisons: PCC between CONUS404 and AORC hourly rain rates drops to about 0.18. CONUS404 also produces excessive hours with very light precipitation, reducing its hourly rainfall detection skill (HSS often <0.4) by lowering the probability of detection and increasing the false-alarm ratio. Overall, the three datasets exhibit complementary strengths and weaknesses. Radar-based MRMS provides the most accurate depiction of extreme rainfall structure and evolution, but its limited temporal coverage makes it less suitable for long-term studies. IMERG-Final and CONUS404 extend the temporal and spatial coverage of high-resolution rainfall information, especially where ground observations are sparse, yet their estimation biases highlight the need for careful application. The results of this study provide a scientific basis for selecting suitable precipitation datasets for extreme rainfall analysis over CONUS and for designing multi-source data-fusion strategies to improve monitoring.