The heaviest rainfall over 61 yr hit Beijing during 21-22 July 2012. Characterized by great rainfall amount and intensity, wide range, and high impact, this record-breaking heavy rainfall caused dozens of deaths and extensive damage. Despite favorable synoptic conditions, operational forecasts underestimated the precipitation amount and were late at predicting the rainfall start time. To gain a better understanding of the performance of mesoscale models, verification of high-resolution forecasts and analyses from the WRF-based BJ-RUCv2.0 model with a horizontal grid spacing of 3 km is carried out. The results show that water vapor is very rich and a quasi-linear precipitation system produces a rather concentrated rain area. Moreover, model forecasts are first verified statistically using equitable threat score and BIAS score. The BJ-RUCv2.0 forecasts under-predict the rainfall with southwestward displacement error and time delay of the extreme precipitation. Further quantitative analysis based on the contiguous rain area method indicates that major errors for total precipitation (> 5 mm h-1) are due to inaccurate precipitation location and pattern, while forecast errors for heavy rainfall (> 20 mm h-1) mainly come from precipitation intensity. Finally, the possible causes for the poor model performance are discussed through diagnosing large-scale circulation and physical parameters (water vapor flux and instability conditions) of the BJ-RUCv2.0 model output.