Abstract: Since initial errors of ensemble members grow toward different directions in atmospheric phase space, within the context of verificationmethods of numerical weather prediction, an index of composite predictability (ICP) for persistent heavy rainfall is established, whichprovides a scientific method for studying predictability and initial error growth. This index (ICP) includes three mathematical models: index ofsingle variable (ISV), index of composite predictability (ICP) and the definition of the ensemble forecasting ability for each member. Verificationof the method is made utilizing T213 global ensemble forecast data for two episodes of persistent heavy rainfall events in Huaihe River basin.The ETS score of rainfall over 10 mm·h-1 and root-mean-square error (RMSE) of 500 hPa geopotential height are selected for ISV evaluation to represent the ability to predict precipitation and atmospheric circulation, respectively. The results show that ICP can pick out the best and the poorest ensemble members effectively, which have a marked discrepancy in predicting the persistent heavy rainfall event. As for predicting the location, duration and the movement speed of the large-scale weather systems, such as blocking high, subtropical high and the East Asian trough, good members make much more accurateness forecast than poor members. The same is true for predicting the evolution of mesoscale systems, such as shear line in low level and Southwest Vortex. There exist relative large deviations for the poor members in fore-casting the large scale circulation, mesoscale systems and the location of rain belt. Preliminary results show that the ICP for persistent heavy rainfall is reasonable and reliable.