The paper presents a new algorithm using curve moment invariants to recognize partially occluded objects in an image. A database is initially built to store the information of a variety of objects after calculating their features represented by curve moment invariants. The images are described by series of corresponding curve moment invariants of the objects as their descriptors. Accordingly, the occluded objects in the images can then be recognized after querying in the model database with prior knowledge of different objects. Results from experiments turn out to be reliable and feasible to recognize images with occlusion.