The present study examined whether obtaining additional articulatory information about the sounds of a difficult second-language contrast (English /ε/-/æ/ for German speakers) could help nonnative listeners to encode a lexical distinction between novel words containing these two categories. Novel words (e.g., tenzer-tandek) were trained with different types of input and their recognition was tested in a visual-world eye-tracking task. In Experiment 1, a baseline group was exposed to the words audio-only during training, whereas another group additionally saw videos of the speaker articulating the target words. In Experiment 2, listeners were asked to repeat the target words themselves as part of their training. It was found that both audiovisual input and word repetition during training resulted in asymmetric fixation patterns at test: Words containing /ε/ were recognized more readily than those with /æ/, mirroring the recognition asymmetry reported for real English words. This asymmetry was not present for the audio-only group, in which target words with the two vowels were fixated similarly. The results suggest that articulatory knowledge, acquired through both passive exposure to visual information (Experiment 1) and active production (Experiment 2), can help distinguishing words with difficult foreign sounds.