Abstract: Focusing on the widespread freezing-rain event in China during February 21–25, 2024, this study evaluated freezing-rain zone identification methods and assessed the applicability of the GraphCast machine-learning forecast system. Using station observations and ERA5 reanalysis, three regional identification approaches—key indicator layer thresholds, cold/warm layer thickness thresholds, and an area-integral comparison of cold/warm layer intensity—were compared within GraphCast forecast fields. In addition, sounding profiles and temperature stratification were analyzed to quantify forecast skill and error sources. Results indicate the following. (1) Among the three approaches, the cold/warm layer thickness threshold yielded the closest agreement wth observed freezing-rain areas. In contrast, the key indicator layer threshold approach severely underestimated the affected zones, while the area-integral comparison method, although capturing meridional displacement of the main freezing rain area, suffered from missed early freezing rain in central Hunan and incorrect late-stage identification in the lower reaches of the Yangtze River. (2) GraphCast reproduced representative sounding structures and the evolution of the low-level cold dome and mid-level warm layer. However, systematic misses occurred over Henan and northern Anhui on February 21 and over western/southern Guizhou on February 24–25, accompanied by false alarms over north–central Jiangxi and Zhejiang. (3) Error analysis revealed systematic low-level temperature biases: an underestimated cold dome on February 21 and an overestimated cold dome combined with an overestimated warm layer on February 24–25. These biases were identified as the primary drivers of forecast-area discrepancies. Overall, the findings identify cold/warm layer thickness thresholds as the preferred approach for GraphCast-based freezing-rain prediction and expose critical deficiencies in low-level temperature and stratification forecasts, providing a basis for refining machine-learning-based forecasts of extreme freezing-rain events in China.