This research outlines the predominant dialogue and performance characteristics of three-person interpreted telephone speech during service-oriented dialogues, in comparison with those of two-person non-interpreted dialogues. An empirical study was conducted in which 12 native English speakers each made one telephone call through an experienced telephone interpreter to a Japanese confederate who did not speak English, and a second call to a Japanese confederate fluent in English. In total, 24 dialogues were collected, each one containing two successfully completed service tasks, or 48 tasks in total. This paper reports on comparisons performed between three-person interpreted and two-person non-interpreted speech, based on the same pool of tasks and English subjects. The unique characteristics of interpreted telephone dialogues are outlined, including structural and referential features, miscommunications and other performance characteristics, confirmatory language and linguistic indirection. In addition, an analysis is presented of interpreters' strategic management of turn shifts, and of the content, sequencing and chunking of information passed among speakers. The long-term goal of this exploratory research is the theoretical account and modeling of human dialogue, the specification of preliminary target requirements for future automatic systems, and the optimization of human performance within those systems.