The STARE (STructured Analysis of the REtina) project uses object-identification and artificial intelligence techniques to provide automated diagnoses from color pictures and fluorescein angiograms of the ocular fundus, or automated change detection from sequential images. As part of the object-identification process, we apply expert judgment and experimentation to define features--such as size, shape, color, and texture--of objects (disk, blood vessels, lesions) in digitized images. In our initial investigations, we explored color alone, because it yields a great deal of information in the classification process. We verified that even similarly colored lesions (exudates, cotton-wool spots, and drusen) could be classified by color with moderate success by a quadratic discriminant function. When color alone is not sufficient, refinement in the classification of objects may be achieved by using more features in statistical pattern recognition. Ultimately, we build a description of the fundus image which can be used either to identify one or more diagnoses that can cause the pattern of lesions in the ocular fundus or to recognize change in sequential images.