Five adult subjects with normal hearing and vision viewed tachistoscopically projected photographs of a talker uttering six phonemes. Experiment 1 determined discrimination of the visemes as a function of exposure duration (12–14 msec) and demonstrated that recognition of certain lip postures was a direct function of duration whereas for other postures duration appeared to interact with other variables. In Experiment 2, fixed duration stimuli (17 msec) were followed immediately by a variable duration masking stimulus (12–45 msec), and in Experiment 3 the test stimuli varied (22—52 msec) and the masking stimulus was fixed (45 reset). Results showed that under both conditions test stimuli were masked when the masker was at least as long as the test stimuli. In Experiment 4, the test stimuli and masking stimulus were held constant (15 msec and 45 msec, respectively) while a variable (7—37 msec) ISI was interposed. Delaying the masker did not improve recognition scores. Conclusions were (a) lip postures are subject to backward recognition masking and the effect varies in degree; (b) the processing of lip postures begins with a short-term storage of the posture; and (c) the initial stage of perceptual processing requires more than 37 msec. To the extent that this task parallels the speechreading process, the results would not support training procedures based at the level of single visemes.