This paper describes a prototype of browser which uses speech recognition to navigate and retrieve information through the World Wide Web. Our multimodal browser shows how it is possible to achieve, with speech, some tasks which are classically executed by the means of the keyboard and/or the mouse. We describe the way we use natural language to help users, novices or experts: with dialogue and sub-dialogue procedures included in an interaction system, we guide the user to a man/machine conversation where he/she can tell his/her needs and goals.