Two experiments were conducted to determine whether listeners' ability to use allophonic variation to identify word boundaries is influenced by speaking rate. Listeners in both experiments were presented two-word sequences (such as great eyes) spoken by naturally fast and naturally slow talkers; in one experiment the sequences were presented in quiet and in the other they were presented in noise. The listeners' task was to identify the intended sequence from among four choices with alternative segmentations (e.g. great eyes, gray ties, great ties, gray eyes). In both experiments performance was worse for the sequences produced by the naturally fast talkers than for those produced by the naturally slow talkers. This finding suggests that the extent to which allophonic variation contributes to the identification of word boundaries may depend on the rate at which the speech was produced.