Theoretical models argue that listeners' perception of second language sounds is heavily influenced by their native language phonology, a prediction borne out by behavioural studies. However, we lack quantitative models capable of making more precise predictions of the way in which the first and second language sound systems interact. The current study introduces a computational modelling framework that permits comparison of different second language learning strategies which vary both in the degree of first language influence as well as in the manner in which second language input is combined with existing first language knowledge. Six different model variants were evaluated by comparison with behavioural data on a task involving the identification of intervocalic consonants of Castilian Spanish by Mandarin Chinese listeners. All approaches demonstrated a similar pattern of rapid improvement with exposure to that observed in listeners. However, approaches that made use of independent first and second language models made the best predictions. An approach that excluded first language influence both predicted lower listener identification levels in the initial stages of learning and higher scores in later stages, demonstrating that first language experience helps to bootstrap second language sound learning but ultimately hinders identification. However, modelling outcomes also demonstrate that no single approach can account for the identification patterns for all consonants, suggesting that learners deploy different approaches to the learning of individual sounds.