With the growth of wind power penetration, the study and application of wind power prediction which could improve security, stability and receptiveness of the grids is of great importance. This paper analyses two prediction methods: one is based on wind speed prediction; another is based on wind power prediction. Then the ANFIS (adaptive neuro-fuzzy inference system) prediction model is established. Taking the real data from a wind farm in the west of Jilin province to do the real-time multi-step rolling wind power prediction under two methods based on ANFIS, and the results of it are compared with the results of the linear regression method, the moving average method and the persistence method. It is shown that the prediction accuracy of the ANFIS method is higher and the availability of ANFIS method can be proved. Further more, it also illustrates that the prediction accuracy of ANFIS method based on wind power prediction is better than ANFIS method based on wind speed prediction.