It is an important task to remove the redundant information in hyperspectral images. In this work, a band selection method for hyperspectral data based on matrix mode (BSMM) is proposed by combining the coefficient of variance and the correlation coefficient and a matrix factor is defined. By using the coefficient of variance and the correlation coefficient, we analyze the separability and correlation of hyperspectral images and compare the standard deviation and coefficient of variation when the measurement of spatial dimension information is discussed. In addition, interval mapping is adopted to remove the difference of variation interval between the coefficient of variation and the correlation coefficient. The efficiency of this method is demonstrated by the analysis of AVIRIS hyperspectral data.