A problem with trying to create voice interfaces to existing HTML web pages is that most web pages are designed and optimized for graphical browsing. Our approach has been to consider voice interfaces to web information not as a replacement for desktop browsing using a graphical interface, but as a way to meet mobile users' needs for specific information. Such an interaction can be more directed and focused in nature than open-ended browsing tasks. In order to create usable voice interfaces to web information, we have identified a set of information structures that are commonly found on web pages and explored how they can be presented and navigated by voice. We have codified these structures in an XML annotation language called VXPL that can be used to annotate existing HTML code to indicate where these information structures exist. VXPL is designed so that it can be used for manual or automatic annotation of HTML code. In addition, we have implemented an automated transcoder that converts VXPL into VoiceXML with increased support for navigation and usability of the resulting interface.