In this paper, I will examine some of the difficulties faced by the linguistic fieldworker who is attempting to observe and record "natural" conversations, and I will reconsider the long-held sociolinguistic notion of the observer's paradox by recasting it within Bell's (1984) framework of audience design theory. Using data gathered during my own fieldwork, I will once again call into question the idea of a single, unmarked, unperformed vernacular, the access to which is supposedly blocked by the observer's paradox. Finally, I will demonstrate that "performed" or "self-conscious" speech produced for the fieldworker can be useful in systematic linguistic analysis, and in gaining insights into local language ideologies and linguistic norms.