The performance of the current generation of automatic speech recognition hardware is generally poorer than that of human listeners. However, the effective recognition accuracy of voice input systems incorporating such devices can be increased by making careful use of interactive dialogues. The criteria for a good dialogue control strategy are discussed, and a report of a trial of a particular scheme is given. Results from this trial indicate that word misrecognition can be decreased by about 25% without compromising the fluency of the man/machine interaction.