This article shows that first language (L1) transfer may not be effectively maintained in the interlanguage due to confounding factors in the second language (L2). When two factors, "A" and "B," are correlated in the L2, second language learners may only acquire "B," even if "A" is present in the L1. Transfer may not be effective because "B," being more robust in the input, conceals "A." Native speakers, on the other hand, generalize "A" in spite of "B." The variables in question are weight-sensitivity ("A") and positional bias ("B") in English, both of which can predict the location of stress in the language. I show that two seemingly target-like groups of second language learners of English (speakers of Mandarin and speakers Portuguese) fail to accurately generalize weight-sensitivity in the language, and instead display response patterns which are predictable given the existing positional bias in English stress.