Abstract: Due to the residual circulation of typhoon Doksuri (2305), heavy rainstorm events occurred in North China from July 29 to August 1, 2023. This study evaluates the performance of three operational models at varying forecast lead times: EC-IFS (the Integrated Forecast System by European Centre for Medium-Range Weather Forecasts), NCEP-GFS (Global Forecast System by the National Centers for Environmental Prediction), and CMA-GFS (China Meteorological Administration’s Global Forecast System). The possible causes of the forecast biases are also analyzed. Results show that (1) the forecast capability of the three models decreases with an increasing forecast lead time. Among all, EC-IFS outperforms the other two models with an averaged spatial correlation coefficient (SCC) of >0.5 at lead times of 24−120 h. The SCC of NCEP-GFS is similar to that of EC-IFS with lead times of 24−48 h, but it significantly decreases and fluctuates after 72 h, whereas the SCC of CMA-GFS decreases to negative values. (2) In terms of rainfall location, EC-IFS well reproduces the distribution pattern of the heavy rain, while the forecasted rainfall in NCEP-GFS is situated to the south of the heavy rain due to the eastward location of the forecasted subtropical high and weakened south flow. CMA-GFS predicts rainfall noticeably eastward, mainly due to the shifted subtropical high and continental high, allowing the residual vortex to move northward. With respect to intensity, all three models commonly underestimate the extreme value of heavy precipitation. (3) Vertical structures of diabatic heating indicate that the differences in model performance are mainly attributable to the uncertainties of parameterization for model physical processes. Almost all models are unable to accurately simulate the observed precipitation-relative humidity tilting structure, indicating that models have deficiencies in parameterizing sub-grid physical processes such as convection and clouds, which are the main causes of model forecast bias.