We propose an approach to texture characterization and comparison that directly uses the information of digital images of the earth surface without requesting a prior distinction of structural 'patches'. Digital images are partitioned into square 'windows' that define the scale of the analysis and which are submitted to the two-dimensional Fourier transform for extraction of a simplified textural characterization (in terms of coarseness) via the computation of a 'radial' power spectrum. Spectra computed from many images of the same size are systematically compared by means of a principal component analysis (PCA), which provides an ordination along a limited number of coarseness vs. fineness gradients. As an illustration, we applied this approach to digitized panchromatic air photos depicting various types of land cover in a semiarid landscape of northern Cameroon. We performed 'textural ordinations' at several scales by using square windows with sides ranging from 120 m to 1 km. At all scales, we found two coarseness gradients (PCA axes) based on the relative importance in the spectrum of large (> 50 km [exponent) -1), intermediate (30-50 km [exponent)-1), small (10-25 km [exponent) -1) and very small (<10 km [exponent) -1) spatial frequencies. Textural ordination based on Fourier spectra provides a powerful and consistent framework to identifying prominent scales of landscape patterns and to compare scaling properties across landscapes.