Consonant confusion matrices were obtained for four sets of CV and VC nonsense syllables presented both in quiet and in the presence of a masking noise. A sequential method of partitioning transmitted information for confusion matrices was developed and used to test the hypothesis that when the internal redundancy of feature systems is taken into account, certain articulatory and phonological features of consonants consistently account for transmitted information better than other, closely related, features. Results of the analyses indicate that for most confusion matrices several feature systems can be shown to account equally well for transmitted information, and that across syllable sets and listening conditions, there is little consistency in the identification of perceptually important features. The implication of these findings with respect to the existence of natural perceptual features for consonants is discussed.