A method called “wavelet packet approach to sub-band decomposition independent component analysis” was introduced to unmix hyperspectral mixed pixels. Using wavelet packet decomposition technique to improve independent component analysis, a blind unmixing method for hyperspectral data was proposed considering the characteristics of hyperspectral data, and this method could overcome the drawback derived from statistical independence assumption in independent component analysis. Two groups of synthetic data and three groups of indoor data were used to evaluate this method. The experiments show that spectra and fractional abundances of the endmemebers can be retrieved precisely using the proposed method, and its accuracy is significantly higher than that of independent component analysis. This approach is useful for blind unmixing of hyperspectral remotely sensed imagery.