The technique of "voiceprint identification" has been invested with a myth of infallibility, largely by means of a specious analogy with fingerprints. The refusal of its chief proponent to submit to a properly controlled test of his ability, coupled with the inability of observers in independent studies to get comparably low error rates, is sufficient basis for skepticism. There is reason furthermore to doubt whether spectrograms could ever form a valid basis for absolute identification. The present study replicated "The 'Voiceprint' Game" with even more stringent constraints on selection and arrangement of spectrograms representing three undisguised tokens of the same word or phrase by each of three adult male native speakers of North Midland American English. The results show that (1) most of the similarity between any two spectrograms of the same word is "linguistic," arising from that particular word's being spoken in the same or similar dialect and context; and (2) differences between the spectrograms of the same word spoken by two different speakers of similar dialects are "small" in comparison to within-speaker variations for that word. The "Voiceprinter" is more properly likened to the "lie detector" in that stable guidelines for interpreting its graphic records appear permanently beyond reach. (Author/AMM)