Hyperion images have the features of high spectral resolution , large quantity of data, strong correlation among adjacent bands and high information redundancy. These features have brought many problems to the data processing and interpretation of them. To solve these problems, an idea of band selection and dimensionality reduction by combining the segmented principal component analysis with the band index algorithm is proposed. The adaptive band selection method, band index method and cumulative contribution rate method are used to carry out a comparative study in band selection. The results obtained by four band selection methods are analyzed in optimal band combination, feature reparability and image transformation. The experimental results show that the segmented principal component analysis and band index algorithm can not only restrain the phenomenon that some local important spectra may be filtered due to the global transformation, but also can take into account the adaptive partition subinterval correlation among the bands and effectively reduce the hyperspectral data dimension. The method is better than the traditional adaptive band selection method, band index method and cumulative contribution rate method in band selection.