The use of computers in architectural design has shifted from what initially was a very numerical approach into a graphical way and thus into a designerly way of working. Through the extensive use of WIMPs CAAD packages today are more or less user-friendly. In the meantime the ever increasing functionality of those packages inflate the number of commands, actions, options and choices which have to be activated with the mouse in order to generate a structured drawing. As a result packages become rather cumbersome to handle. In this research we have explored a new medium, namely speech, to tell the machine what to do. With software for speech recognition and making use of a head-mounted microphone we have built a far more user-friendly way of handling a CAD package. Grids, snap, ortho, layers, settings and other commands that can be used in combination with mouse actions are activated and deactivated by voice command. We are convinced that this is a step further towards a more easy and natural way of using computers while designing and certainly a way for speeding up the modelling of real architecture. The experiment has been conducted for AutoCAD with the software for speech recognition by Lernout & Hauspie. This new way of working is not limited to architecture and the principle can be used right now for any other package, provided it has a programming interface.