This paper considers the issue of non-target language representations analysing data from the acquisition of novel English synthetic compounds by bilingual (Basque-Spanish) speakers. Past research indicated that English L1 acquirers do not include regular -s inflection within compounds, whereas they do include irregulars (*"clawsmarks" versus "teethmarks") and claimed that this was an innate constraint. More recent research contradicts this claim in the case of L2 acquirers. The aim of this study is to report on further work investigating this issue with bilingual learners in an EFL setting. The results indicate that there is a significant contrast between native and non-native speakers of English in novel compound production and also show that there is a significant effect on the proficiency level of the three groups tested. These results are discussed in the light of the predictions made by the level-ordered morphology model and the dual-mechanism model and also consider the role of the learners' L1s and the amount of target language input in determining novel compounding in L3 English. (Contains 8 tables, 3 notes, and 2 footnotes.)