The effectiveness of retrieval practice in supporting learning can be enhanced if retrieval cues are varied across learning cycles. This has been previously demonstrated in a study examining foreign vocabulary acquisition and contrasting varied and constant sentences with embedded to-be-learned foreign words, where sentences were used to prompt retrieval of the meanings of those words. However, to date the benefits of variable over constant retrieval were assessed solely in an explicit test of the meanings of the learned foreign words. In contrast, the present study assessed whether variable retrieval yields benefits over constant retrieval also in transfer tests that require the understanding of the meanings of foreign words but do not ask for those meanings directly. In a series of experiments, this was assessed with the use of a semantic relatedness task, which required speeded decisions as to whether foreign words were semantically related or unrelated to the preceding primes, and a sentence verification task, which required decisions as to whether foreign words fit into semantic contexts of novel sentences. The obtained results confirmed the superiority of learning via varied over constant retrieval practice also for the transfer tasks, suggesting that varied retrieval practice leads to effective embedding of the learned words in semantic knowledge structures.