We present an attractive methodology for the compression of facial gestures that can be used to drive interaction in real time applications. Using the eigenface method we build compact representation spaces for a variety of facial gestures. These compact spaces are the so called eigenspaces. We do real time tracking and segmentation of facial features from video images and then use the eigenspaces to find compact descriptors of the segmented features. We use the system for an avatar videoconference application where we achieve real time interactivity with very limited bandwidth requirements. The system can also be used as a hands free man-machine interface.