Infants' sensitivity to word units in fluent speech was examined by inserting 1 sec pauses either at boundaries between successive words (Coincident versions) or between syllables within words (Noncoincident versions). In Experiment 1, 24 11-month-olds listened significantly longer to the Coincident versions. In Experiment 2, 24 four-and-a-half-and 24 nine-month-olds did not exhibit the preference for the Coincident versions that the 11-month-olds showed. When the stimuli were low-pass filtered in Experiment 3, 24 11-month-olds showed no preference for the Coincident versions, suggesting they rely on more than prosodic cues. New stimulus materials in Experiment 4 indicated that responses by 24 11-month-olds to the Coincident and Noncoindent versions did not depend solely on prior familiarity with the targets. Two groups of 30 11-month-olds tested in Experiment 5 were as sensitive to boundaries for Strong/Weak words as for Weak/Strong words. Taken together, the results suggest that, by 11 months, infants are sensitive to word boundaries in fluent speech, and that this sensitivity depends on more than just prosodic information or prior knowledge of the words.