The effect of repeated exposure to prevoiced and voiced stimuli on the voiced/voiceless category boundary was compared for monolingual native English speakers and bilingual native speakers of Thai. The English speakers, for whom the two adaptors were members of the same phoneme class, showed equivalent shifts towards the adapting stimulus under both conditions. The Thai subjects, who perceived the two adaptors as belonging to two distinct phoneme classes, showed a shift of the voiced/voiceless boundary only for the voiced adaptation condition; exposure to the prevoiced adaptor had no effect. Since Eimas and Corbit (1973) first demonstrated selective adaptation for the linguistic feature of voicing, their technique has been used with other categories and classes of speech (Cooper & Blumstein, 1974; Eimas, Cooper, & Corbit, 1973; Morse, Kass, & Turkienicz, 1976). The procedure is simple: repeti­ tion of a token taken from one end of a linguistic continuum leads to a shift in the category boundary, towards the adapting stimulus. In the original demonstration, subjects adapted with a Iba/ found stimuli previously lying on the Iba-pal boundary to be more Ipal-like, while those exposed to Ipa/ re­ ported just the reverse. The first studies of selective adaptation used as adaptors stimuli chosen from the extreme ends of the continuum, stimuli which could be considered "good" exemplars of their phoneme class. It was, therefore, of both empirical and theoretical interest to test the effect of adaptors closer to the category boundary, which, although they were not as "good" exemplars acoustically, would still be unambiguously perceived as belonging to one or the other phoneme class. Using the voice onset time (YOT) continuum, ,Anderson (1976) and Miller (Note 1) found that stimuli close to the preadaptation category boundary produced less adaptation than more extreme stimuli. Miller (Note 1) interpreted these results to support the existence of feature detectors selectively tuned to overlapping distributions of YOT values. The closer the stimulus to the optimal values, the more strongly the feature detector fires, and the greater the This research was funded by Grant R03 MH 26168 from NIMH and FRAP 11184 from the Research Foundation of the City University of New York. The author wishes to thank Drs. F. S. Cooper and A. M. Liberman for making available the facilities of Haskins Laboratories, Ralph Bianca for gathering the data, and the Thai Student Association of Columbia University for providing volunteers. Above all, thanks are due to the Thai students themselves, who endured not only the experimental sessions, but a 3-h round-trip subway ride as well. 347