Statistical properties of the chromatic spectrum of landscapes are studied by Karhunen-Loève (K.L.) transform. The information is found to be compressed into a few dominant eigenvectors of the covariance matrix of multispectral data. Natural and man-made objects are shown to differ by their covariance and therefore by the distribution of their eigenvalues. Feature selection is performed by using the first eigenvector as a chromatic filter. The respective influences of three elements of the landscapes considered (i.e. vegetation, sky, and cars in a parking lot) are assessed. Further applications to the automatic classification of the content of landscapes are discussed, and a hypothesis is proposed for the origin of the chromatic response of the human eye.