The mismatch negativity (MMN) auditory evoked potential reflects detection of minimal stimulus differences at a pre-attentive level. It has been reported that the MMN reflects acoustic, rather than phonetic, processing of speech stimuli [Sharma et al., Electroenceph. and Clin. Neurophys. 88, 64--71 (1993)]. If this is true, then every stimulus pair from a continuum along which the acoustic differences between each step are equivalent should elicit an equivalent MMN, even when the acoustic differences are not perceptible to the listener (e.g., within-category differences). In this study, stimuli were two-step syllable pairs taken from a nine-step synthetic continuum varying in place of articulation from /da/ to /ga/. Results of conventional labeling and discrimination tasks demonstrated that the stimuli were categorically perceived. The MMN was tested using three stimulus pairs: a pair in which syllables were not perceptibly different (items seven and nine, a within-category distinction), the pair best discriminated (behaviorally) by each individual subject (a between-category distinction), and a pair comprising the endpoints of the continuum (items one and nine). Results are discussed in terms of their implications for using the MMN as an objective measure of speech discrimination ability and as a tool for studying aspects of speech perception.