Considering the high degree of correlation in the visible spectrum, the principal wavelengths from spectral measurements of radiance recorded in spectral images were selected using a method based on principal components analysis (PCA). It seems to be that this is the first time that, instead of using spectra, data is taken directly from the “slices” of spectral images; the method has the advantage of preserving the structure of the original data in the reduced data set. A “true” dimensionality of five wavelengths resulted for all the analyzed images. The averages of the selected wavelengths for 10 spectral images produced good results for a human observer. These results were possible using only four wavelengths. Though PCA by itself is not able to include the impact of specific sensors on the selection of basis functions, results suggest that the variable selection method used in this work (which is not just PCA) yielded objective information of the structure of the physical stimuli (i.e., the spectral structures) that have been shaping the visual systems of animals and insects since many years ago.