Land cover classification is one of the important applications of satellite images. The accuracy of the classification process depends on the feature selection. In multispectral satellite images, the separability of the features depends on the band combinations used. This work demonstrates the change in the accuracy of the classifiers with different band combinations and different distance measures used for analyzing the separability. Landsat-8 images have been classified into four land cover classes using the Maximum Likelihood (ML) Classifier, Minimum Distance (MD) Classifier, and the Spectral Angle (SA) mapper. The spectral separability between each of the land cover classes is analyzed for the band combinations of 3-2, 4-3-2, 3-2-1, 7-6-5-4, 7-6-5-4-3, using the Jeffries-Matusita Distance measure, the Euclidean Distance, and the spectral angle measure. It is shown that maximum separability and hence the optimal accuracy of 85.81% is obtained with a SA mapper using the spectral angle measure on a three-band combination of 4-3-2. An accuracy of 80.12% is achieved with an ML classifier using Jeffries-Matusita distance measure with a band combination of 4-3-2. Lastly, the MD classifier gives an accuracy of 76.56% using the Euclidean distance measure with a band combination of 4-3-2.