This study investigates the compensation for assimilated sounds in a first and second language, and the mental representation of phonological alternations, using evidence from voicing and devoicing assimilation in French. Previous research established that listeners can use abstract phonological knowledge of alternations in their L1 to compensate for assimilations during word recognition (Gaskell & Marslen-Wilson, 1996, Darcy et al., in press). It also provided evidence that highly proficient late learners had acquired the ability to compensate perceptually for the assimilatory pattern of their L2 (Darcy et al., 2007), suggesting that they developed a comparable phonological knowledge of alternations for L2. The present study explores more in depth the representation of such knowledge in a first and a second language. First, sections 2 and 3 provide acoustic analyses of stimuli to ensure that assimilations are categorical, followed by a confirmation that stimuli are also perceived categorically (identification task). Section 4 evidences an asymmetry in compensation between voicing and devoicing for native listeners, suggesting a difference in the mental representation of voicing vs. devoicing. Non native listeners (English learners of French) show in section 5 a different response to the same stimuli, suggesting that they might use a different processing mechanism or build less detailed representations than French listeners (Sebastian-Gallés et al., 2006). The nature of those possible mechanisms and representations is discussed.