Phonetic convergence is linguistically and socially selective. The current study examined the constraints on this selectivity in convergence to Southern American English by non-Southern Americans in a word shadowing task. Participants were asked either to repeat the words after the model talker, to repeat the words after the model talker from Louisville, KY, or to imitate the way the model talker from Louisville, KY, said the words, in a between-subject design. Acoustic analysis of the participants' productions revealed significant phonetic convergence on word duration and back vowel fronting, but not on /aɪ/ monophthongization, across all three instruction conditions. These findings suggest social selectivity such that convergence on stereotyped variants is avoided, but convergence to a talker with a non-prestigious variety is not. A perceptual assessment of convergence confirmed the acoustic results, but also revealed significantly more convergence in the explicit imitation condition than in the two repetition conditions. These findings suggest that explicit instructions to imitate lead to greater convergence overall, but do not completely override social selectivity. A comparison of the acoustic and perceptual assessments of convergence indicates that they provide complementary insights into specific features and holistic patterns of convergence, respectively.