Abstract: To better understand the forecasting performance of the China Meteorological Administration Mesoscale Model (CMA-MESO) and the Zhejiang WRF-ADAS Rapid Refresh System (ZJ3KM) for short-term heavy precipitation in Zhejiang Province, this study systematically assesses the two models using precipitation process verification, traditional verification methods, as well as FSS (Fractions Skill Score) and MODE (Method for Object-Based Diagnostic Evaluation). The analysis is based on short-term heavy precipitation events associated with five typical synoptic patterns—cold shear, warm shear, upper-level trough, tropical cyclone, and subtropical high—during June–September 2023. Model performance is assessed in terms of rainfall intensity, spatial distribution, diurnal cycle, and structural characteristics.The results indicate: (1) CMA-MESO performs better in capturing the rainfall coverage and frequency for warm-shear and tropical cyclone types, whereas ZJ3KM exhibits superior skill in simulating the total precipitation, frequency, and intensity distributions for cold-shear and subtropical high types. Both models show pronounced biases for upper-level trough cases. (2) Regarding diurnal variation, CMA-MESO generally matches observations but underestimates precipitation in the subtropical high type, whereas ZJ3KM tends to overestimate precipitation frequency and advance the timing of peak rainfall. (3) CMA-MESO achieves a higher Threat Score (TS) and lower false alarm rate for the tropical cyclone type, whereas ZJ3KM performs better for the subtropical high type and achieves higher hit rates and smaller domain biases for most precipitation types. (4) Spatial verification indicates that CMA-MESO performs better for tropical cyclone, while ZJ3KM provides superior precipitation intensity indication for other types, albeit with location biases. Both models show insufficient identification of heavy precipitation objects. Regarding spatial matching, CMA-MESO demonstrates better performance for cold- and warm-shear cases, whereas ZJ3KM excels in the remaining three categories. For position bias, except in cold shear, CMA-MESO tends to have eastward and southward deviations, while ZJ3KM shows eastward and northward biases. Overall, optimal operational application should integrate the strengths of both models--leveraging CMA-MESO's trend evolution information, and ZJ3KM's intensity and local small-scale details, while applying targeted corrections to mitigate their respective location biases.