The present study addresses a long-standing assumption in the field of applied linguistics: that presenting new second language (L2) vocabulary in semantically grouped sets is an effective method of teaching. Participants learned 32 new L2 labels for concepts from four different semantic categories in either a related or unrelated condition. At test, participants translated words in both translation directions. We found a semantic interference effect both during the encoding of information into memory and during the retrieval of information in translation. We discuss these findings in terms of theoretical models of L2 lexical representation and development, as well as in more practical terms of L2 curriculum design and vocabulary instruction.