Most of the multispectral sensors acquire data in several broad wavelength bands and are capable of extracting different Land Cover features while hyperspectral sensors contain ample spectral data in narrow bandwidth (10- 20nm). The spectrally rich data enable the extraction of useful quantitative information from earth surface features. Endmembers are the pure spectral components extracted from the remote sensing datasets. Most approaches for Endmember extraction (EME) are manual and have been designed from a spectroscopic viewpoint, thus neglecting the spatial arrangement of the pixels. Therefore, EME techniques which can consider both spectral and spatial aspects are required to find more accurate Endmembers for Subpixel classification. Multispectral (EO-1 ALI and Landsat 8 OLI) and Hyperspectral (EO-1 Hyperion) datasets of Udaipur region, Rajasthan is used in this study. All the above mentioned datasets are preprocessed and converted to surface reflectance using Fast Line-of-sight Atmospheric Analysis of Spectral Hypercube (FLAASH). Further Automated Endmember extraction and Subpixel classification is carried out using Multiple Endmember Spectral Mixture Analysis (MESMA). Endmembers are selected from spectral libraries to be given as input to MESMA. To optimize these spectral libraries three techniques are deployed i.e. Count based Endmember selection (CoB), Endmember Average RMSE (EAR) and Minimum Average Spectral Angle (MASA) for endmember selection. Further identified endmembers are used for classifying multispectral and hyperspectral data using MESMA and SAM. It was observed from the obtained classified results that diverse features, spread over a pixel, which are spectrally same are well classified by MESMA whereas SAM was unable to do so.