Ensemble prediction is an effective approach to accounting for uncertainties of initial conditions and model error. By combining the predictability of extended-range, both predictable components and unpredictable random components with different characteristics are treated with different ensemble prediction schemes and strategies. A new predictability-based extended-range ensemble prediction method (PBEP) is proposed. In this method, for predictable component, the uncertainty of model is taken into account through the use of multiple error correction scheme; while the random component probability distribution is obtained from the climate probability distribution of historical data, for the sake of avoiding the influence of model error. Prediction results show that the ensemble prediction method can improve the forecast skill in all regions of the world, and the extents of improvement are different for waves with different spatial scales compared with the operational dynamical extended-range ensemble prediction system of NCC/CMA, exhibiting its potential application perspective to operational extended-range prediction.