Motivated by ESL (English as a Second Language) concerns, this study compares the language of a U.S. situation comedy, Friends, with natural conversation. A corpus of transcripts of the television show and the American conversation subcorpus of the Longman Grammar Corpus are used for analysis. This data-driven investigation combines multidimensional (MD) methodology (Biber 1988) with a frequency-based analysis of a large number of linguistic features associated with the typical characteristics of face-to-face conversation. The results of the MD analysis indicate that Friends shares the core linguistic characteristics of face-to-face conversation, thus constituting a fairly accurate representation of natural conversation for ESL purposes. However, a closer look at the linguistic features revealed interesting functional differences between the two corpora. These differences pointed to distinct functional patterns (e.g., vagueness, emotional language) suggested by the association of linguistic features sharing similar discourse functions.