Spectral feature fitting (SFF) algorithm has been frequently used since 1990s. A modified spectral feature fitting method is introduced here, which can solve some drawback of the general algorithm. The method mentioned here combines SFF with user-defined constraints in spectral absorption feature to extract more accurate target information from hyperspectral image. Two experiments are presented herein, in which three algorithms are used to obtain mineral information from hyperspectral data with different space resolution and SNR. Muscovite, calcite and chlorite etc. are extracted by general SFF, modified SFF and spectral angle mapping (SAM) respectively, and the result indicates that modified SFF algorithm is more effective in differentiating subtle spectral feature and obtaining accurate mineral information. The experiments also demonstrate that the algorithm mentioned here is validated in mineral information extraction.