A heavy rainfall is simulated by GRAPESMESO model， occurring in Huanghuai area on 22 July 2008 to study the forecast uncertainty and the initial perturbation method when establishing GRAPESMESO ensemble prediction system (called GRAPESMEPS) based on TIGGE data. The results suggest that the initial perturbation structure can reflect some information of initial uncertainty. The GRAPESMEPS has the ability to capture the extreme rainfall event and can significantly improve the forecast skill of heavy rainfall. The ensemble mean can capture some characteristics of the mesoscale heavy rainfall validly. The probability forecast gives the high probability area, where the heavy rainfall occurred actually. The ensemble verification results indicate that the relation between the spread and RMSE demonstrates the rationality of the GRAPESMEPS during the early integral period, but the impact of the initial perturbation decreases in the subsequent integral period owing to the model dynamic adjustment.