The suitability of face recognition was investigated as a biometric-based identifier for sheep using a holistic analysis of face images by the independent components technique. Algorithm training was performed independently on several normalized face images from 50 sheep (sets of two, three, and four training images per sheep). The performance of this technique was assessed on a separate set of images (three normalized face images per sheep) using the cosine distance classifier. When 180 to 200 components were extracted, the recognition rate was as high as 95.3% to 96%. As expected, fewer independent components reduced the recognition rate, while a higher number of training images per sheep improved it. Although our results have demonstrated the potential of face recognition as a non-invasive, inexpensive, and accurate novel biometric identifier of sheep, further work should aim at improving recognition rates on a larger set of sheep faces. © 2007 American Society of Agricultural and Biological Engineers.