Reaction time measurements to tokens composed by means of categorial phonetic mismatches have indicated that information which is distributed throughout a syllable is used in making segment identification judgments [Whalen, Percept. Psychophys.35, 49–64 (1984)]. This raises the possibility that speech perception proceeds from syllable perception “backwards” to segment identification. The mismatches created using tokens similar to those of Whalen's experiment 1 are not identical in their demands on the listener. Thus the mismatched fricative may be closer or further from the expected fricative depending on the degree of coarticulatory rounding. For instance, if the formant transitions lead the subject to expect a palatal fricative, the fricative noise of [us]will be closer to the expected fricative than that of [is]. It is proposed in this paper that the perception by syllables approach would predict that subjects would recover from the mismatch more quickly when the actual fricative noise is close to the expected fricative while the perception by segments would predict the opposite. The results of an experiment testing these hypotheses support the perception by segments approach. The interaction (in an ANOVA) between actual fricative and original vocalic context of the fricative was significant [F(3,42) = 3.52,p<0.05] in the appropriate direction.