In this paper I argue that the correspondence approach to stress-epenthesis interactions provides a better match with the data than the multi-level approach. The general argument of the paper is that disruption of normal stress patterns by epenthetic material is caused by one of two factors: avoidance of epenthetic material in prominent positions, and maximization of paradigmatic contrasts. In Selayarese loanwords, the main stress foot is constructed to avoid inclusion of epenthetic vowels anywhere in the foot, while in North Kyungsang Korean loanwords and in Winnebago native vocabulary, epenthetic vowels are avoided in the head position of a foot. In Iraqi Arabic, the invisibility of inserted vowels is motivated by the maximization of contrast between stems of different grammatical types.