A new image recognition algorithm—eigenspace-based approach is proposed. Firstly, recognition system obtains the possible object′s images. Then, all images and image set of each object are compressed to obtain low\|dimensional subspace, called the universal eigenspace and object eigenspace separately. Given an unknown input image, the recognition system projects the image into two kinds of eigenspace. The exact position of the projection in the eigenspace and the distribution formula of samples′ projections determine the object′s species and poses in the image. Experiments are conducted using several objects with complex appearance characteristics. The results show that the proposed approach could keep robustness for the variable poses of object, and could restrain illumination variety.