A short-term wind power prediction method based on similar set of basis time meteorologyis proposed in this paper. Firstly, the historical meteorological records are selected into a set (here called similar set) by grey relational degree, to obviously expose wind power varying law around the basis time. Then the meteorological factors of historical records in the similar set are reduced in dimension by factor analysis method that produces independent factors, this process eliminates correlation between primary meteorological factors and decreases relation nonlinearity between cause and effect. At last, the mapping function from independent factors to wind power is built based on radial basis function neural network (RBFNN) to realize the prediction of wind power. The proposed method is tested by an actual wind farm. Testing results show: the accuracy of wind power predicted by the proposed method is higher than that by the principal-component-based RBFNN method, and much higher than that by the RBFNN method.