The entextualization and recontextualization of speech via transcription is a fundamental methodology of discourse analysis. However, particularly for researchers concerned with sociopolitical issues in discourse, transcription is not a straightforward tool but a highly problematic yet necessary form of linguistic representation. Recent commentators have critiqued the inconsistency of researcher transcripts; by contrast, this article seeks to understand rather than remedy such variability, conceptualizing diversity in transcripts as a kind of linguistic variation. Examining four different types of variation in transcription practice — global format choices based on an analytic focus on content versus form, variation in the details of transcription format in reproducing one's own or others' transcripts, orthographic variation in a single transcript, and variation in translation — the article argues that although reflection about the transcription process cannot overcome the difficulties inherent in this methodology, it can allow scholars to be more attentive to their own transcription choices and their limitations and to make these explicit in their writing.