In this paper, four minerals in the Cuprite data, high-Al muscovite, med-Al muscovite, Calcite and Chlorite, are detected by the MSCPE_BSE algorithm (Mean Square Cross Prediction Error-Based Blind Source Extraction). Firstly, the estimated signals of four minerals are extracted from neighbored pixels according to the AR parameters of their reflectance spectra. Secondly, the correlation coefficient (CC) between the extracted signal and the mineral spectra is treated as a detection value to make a judgment. Compared with Clark's mapping, the experiment result shows that MSCPE_BSE method highlights the desired signal and suppresses the backgrounds. It is able to detect the distribution of the target approximately.