Phonetic categories have a graded internal structure, with some members perceived as better exemplars than others. In a project focusing on voiceless stop consonants, the relation between this structure and processing speed has been examined. Speech continua were constructed in which the voice-onset-time (VOT) value of a word-initial consonant systematically changed from very short to very long. Unspeeded identification tests confirmed that stimuli with very short VOT values are perceived as voiced, whereas those with longer VOT values are perceived as voiceless. Category-goodness rating tests further confirmed that only stimuli within a limit range along each continuum are perceived as good exemplars of the voiceless category; stimuli with shorter or longer VOT values are perceived as poorer exemplars. Initial results from an ongoing experiment involving speeded identification indicate that (1) compared to the best exemplars of the voiceless category, identification of poor exemplars in the voiced–voiceless boundary region is relatively slow, but (2) across a wide range of values, poor exemplars with VOT values longer than the best exemplars are identified as quickly as the best exemplars themselves. These initial findings indicate that internal category structure is not necessarily reflected in processing speed. [Work supported by NIH.]