The three Lisker—Abramson CV series (ba‐pa, da‐ta, ga‐ka) were computer synthesized at C.I.D. in 1‐msec steps. For each 84‐trial block of a high‐uncertainty, same‐different task (no feedback), one of the three continua was chosen at random. On each trial, the ability of three naive listeners to resolve for differences among VOT pairs ranging from VOT = 10–80 msec, in 10‐msec steps was tested. As expected, there was a local maximum performance for each of the three speech continua, thus satisfying one criterion for categorical perception. A minimum‐uncertainty AX task (with feedback) was run with three highly trained subjects for the ga‐ka continuum. Only one standard VOT was presented in a block of trials. Results were not categorical: the minimum detectable ΔVOT was less than 2 msec at VOT = 10 msec and increased as VOT increased. Our results indicate that trained listeners can utilize subtle stimulus cues in synthetic speech if stimulus uncertainty is minimized. [Supported by a grant from NINCDS.]