Based on a few bands and unabundant spectral information of TM remote sensing image, two endmember extraction algorithms are put forward. First, spatial split endmember extraction algorithm, which firstly browses the image, based on the complexity of objects, divides the image into different blocks, then uses hourglass algorithm to extract endmembers. Second, region continuity algorithm, also based on dividing-into-blocks idea, which uses extraction and classification of homogenous object algorithm and spectral correlation energy level matching algorithm to extract endmembers. Finally, comparing the two algorithms, spatial split endmember extraction algorithm runs fast, with little prior knowledge, however, the probability of error extraction endmembers exists; and region continuity algorithm’s precision is higher, needs for prior knowledge, and the segment process is slow. Experimental results show that both spatial-and-spectral combined endmember extraction algorithms can effectively solve the large regional scale, multispectral endmember extraction problem, and have broad application prospects.