Previous research has shown that experiences that match a learner’s prior knowledge (‘schemas’) are not only better retained, but may even use a different memory route than those unconnected to prior knowledge. In particular, such ‘schema-consistent’ content might be learned with less episodic memory involvement, and be instead encoded directly into semantic memory. We tested this claim for word learning in a new language. Dutch participants learned words in Mandarin (Exp. 1), Italian (Exp. 2) or Swedish (Exp. 3). The degree and kind of relevant prior knowledge varied across all three experiments. In Exp. 1, the words’ phonemes either also existed in the participants’ native language Dutch or not. In Exp. 2 and 3, words were either cognates (i.e. overlapping in form and meaning) with Dutch or not, with varying degrees of predictability of the cognate form given the Dutch word. As an established indicator of episodic memory involvement, we asked participants to indicate the voice (male / female) by which a word had been spoken during learning (source memory). The results of the first two experiments showed, first, that schema-consistent items (cognates, or phonetically ‘easy’ words) were recalled better than inconsistent ones. Critically, though, source memory followed the same pattern, contrary to what the ‘schema’ account predicts, and more in line with a competing view that stresses the role of attentional resources. The data for Exp. 3, in which the ‘schema’ is implemented in its strongest form, is currently being collected and will be included.