Deliberate vocabulary learning is common in the L2, however, questions remain about most efficient and effective forms of this learning approach. Bilingual models of L2 word learning and processing can be used to make predictions about outcomes of learning new vocabulary from bilingual (L2–L1) flashcards, and these predictions can be tested experimentally. In the present study, 41 late adult German–English bilinguals learned 48 English pseudowords using bilingual flashcards. Quality of component lexical representations established for the studied items was probed using form priming and semantic priming. The results show that, although all participants were able to establish robust orthographic representations of the studied items, only bilinguals with large L2 vocabularies established high-quality lexical semantic representations. With neither the Revised Hierarchical Model (RHM) nor the Sense Model able to fully account for these findings, an alternative explanation based on a distributed semantic features view of word learning is proposed. Learning implications of the findings are discussed.