In order to improve the classification accuracy of hyperspectral image data, reduce dependence on large number of data sets, an improved method was proposed for feature extraction of hyperspectral data based on the weighted fuzzy C means algorithm. The approach is an extension of previous approach-prototype space feature extraction. Each feature with different weights in terms of weighted fuzzy C means algorithm to ensure the features contain more information after extracted. Experiment results show that compared to results obtained from approach prototype spatial feature extraction method, this method has a stability of data set and higher classification accuracy when extracted a small number of features, which greatly reduces the dependence on the number of data sets of samples, and improves the efficiency of the prototype spatial characteristics method.