ABSTRACT The segmentation strategies used by native and non-native listeners of French were examined in two phoneme-monitoring experiments which required the subjects to detect the presence of word-initial /t/ in potential liaison phrases (e.g., excellent tableau/excellent acteur ) and in non-liaison phrases (e.g., vrai tableau/vrai acteur ). The essentially faultless performance of the natives suggested that the optimal segmentation routine in such phrases is primarily based on the identification of the critical word and, to a lesser extent, on the contextual information, which was more efficiently used to check the outcome of word recognition. In contrast, non-natives tended to rely on guessing strategies, not based on contextual information (contrary to the widely held language teaching recommendation), but on an incomplete acoustic–phonetic/lexical analysis of the signal.