This paper reports a Wizard of Oz simulation of a spoken dialogue system. 100 human-human flight inquiry dialogues were analysed and used as the basis of simulation scenarios. Subjects were given the scenarios and instructed to find the specified information. They were not told anything about the nature of the service but they concluded that they were talking to a computer. 100 simulation dialogues were collected and analyzed. Significant differences were found between the corpora. In general, phenomena in the simulation corpus were more amenable to analysis by computer. Since the only significant difference between the two conditions was the voice quality of the information provider, this suggests that 'natural' quality speech should be avoided in favour of noticeably synthetic speech in dialogue systems. Keywords: dialogue understanding; Wizard of Oz simulation; speech synthesis; Sundial project.