The development of practical methods of evaluating speaker recognizability in processed speech has been hampered by the lack of an adequate system, either physical‐acoustical or perceptual, for classifying voices. Previous efforts to develop taxonomies based on perceived voice traits (PVT) failed for a variety of reasons, including inappropriate scaling procedures and grossly inadequate voice samples. In the present attempt to remedy the situation, more than 500 potential voice descriptors were experimentally screened to yield a reduced set of rating scales with which listeners could reliably describe their perceptions of individual voices. Factor analysis of rating data for a sample of eighty voices revealed at least eight orthogonal PVTs, several of which had not been previously described. Methods of evaluating the effects of speech processing on speaker recognizability from its effects on the reliability and factorial structure of voice rating data are discussed and illustrated.