Recently, a natural and user-friendly human interface has been actively researched. Although conventional user interface is based on keyboards and mice, human body actions can be used for human machine interface. From this point of view, we have been researched into vision-based real-time human motion analysis. The important point is that we have employed vision-based methods, because they do not impose any physical restrictions on users. In this paper, we present a robust hand shape recognition system based on an eigenspace method, in which we employ a simple additional learning feature, and we show its performance through some experiments.