A novel scheme is proposed for the efficient object tracking by using partial projections of a sparse set of pixels to eigenspaces. This paper shows a theoretical framework of the sparse eigentemplate matching and its application to a real-time face tracker. The sparse eigentemplate matching is formalized as a partial projection onto an eigenspace. Only using a small number of pixels, it facilitates an efficient template matching. In the application, a condensation framework is combined with the sparse eigentemplate matching in order to make a robust and efficient tracker. Experimental results show that the condensation tracker can track a face in real time even when the lighting condition changes.