The present study was designed to investigate whether perceptual training in the form of structured identification tasks with synthetic stimuli could have an effect on the perception and the production of voicing contrasts in the stop consonants of Standard French by native speakers of Mandarin Chinese. Twelve adult subjects took part in the experiment, which consisted of a pretest, 3 hours of training in six half-hour sessions, and a posttest. The pretest and the posttest consisted of identifying and imitating natural stimuli, and of identifying synthetic CV stimuli from several continua in which the voice onset time (VOT) duration of the initial stops varied from -60 ms to +130 ms in 10 ms steps. Training materials were 7 sets of synthetic stimuli consisting of a labial stop followed by the vowel /u/, sequenced according to a technique based on perceptual fading. After three hours of training, a) the subjects' identification functions were closer to those of native speakers of French for the trained continuum; b) there had been transfer of the training effect to the other places of consonantal articulation, and to the vowels [a] and [i] for the labial stops; and c) improved performance was also observed at the production level (imitation task). These findings suggest that adults may learn to perceive and produce non-native speech contrasts with limited but structured perceptual training, and that training effect may transfer to phonetic contexts other than those included in the training set. This last result is interpreted as reflecting the fact that although voicing distinctions are actualized at different locations on the VOT continuum in different languages, variations in VOT duration in terms of phonetic context exhibit some well-defined universal tendencies.