Successful word learning requires establishing an initial representation that is sufficiently robust to be retained in memory. Sleep has profound advantages for memory consolidation, but evidence concerning the effects of sleep in young children's word learning is slim and focuses almost exclusively on learning nouns. Verbs are representationally more complex and are often learned from non-concurrent linguistic and observational information (e.g., hearing "let's pour your milk" before the pouring event takes place). What remains unknown is whether initial representations built this way are robust enough to sustain a delay, and how these representations are affected by sleep. We presented two-year-olds with non-concurrent linguistic and observational information about novel verbs and immediately tested their knowledge of the verbs' meanings by evaluating their eye gaze as they looked at potential referents. Then, after a 4-hour delay during which half of the children napped and half remained awake, we retested them to see if they remembered the verbs' meanings. The results demonstrate differences in two-year-olds' representations of a novel verb before and after the delay; specifically, their verb representations withstood the 4-hour delay if they had napped, but decayed if they had remained awake.