This study examines the role of distributional and metrical cues in the learning of an artificial language. French subjects first heard a sample of the language (a continuous string of nonce syllables) and then were required to decide whether bi- or trisyllabic items belonged to the language or not. One group listened to a sample whithout metrical cues, while another group received the same language with metrical segmentation cues - alternation of short and long (accented) syllables.The presence of these metrical cues improved the discovery and the learning of the artificial words. The implications of this research for lexical segmentation and the learning of a new language learning are discussed.