In January 2010, electronic identification of sheep was introduced within the European Union. This article presents an alternative method of verification of sheep identity. Current identification protocols implement a system in which the tag number is traced throughout the traceability system and not the animal; this can lead to misidentification and the possibility of tampering with identifiers. Retinal identification offers a non-invasive, accurate, and tamper-proof method of identification in which the retinal vascular patterns of both eyes are recorded for identification. There are more than 200 distinct breeds of sheep worldwide, so it is important to determine if breed affects the retinal image matching performance as a result of different eye physiology, such as the amount of myelinization (dark pigmentation present on the ocular fundus) that forms within the retinal vascular area that is examined for identification. A previous study determined that the best condition in which to achieve optimum quality retinal images was indoors; therefore, all the retinal images in this experiment were collected indoors. This study sets out the matching score criteria that are required to successfully implement retinal imaging as a viable method of sheep identification. The objectives of this study were to examine the effects of breed and age of sheep on the matching scores recorded and to set a matching score decision criterion for a retinal imaging system. A representative sample (n = 160) of four breeds of sheep (Cheviot, Charollais, Suffolk, and Texel) categorized into two age groups (<2 years and >2 years) were examined. A non-parametric statistical analysis of matching scores showed that there was no observable effect of breed (p = 0.209) or age category (p = 0.181) on the matching performance. The recognition performance of a single-eye retinal system (either the left or right eye retinal image was used for identity) was estimated to have a false match error of 0.43% and a false non-match error of 0.96% at a matching score threshold of 75 when sheep retinal images were obtained indoors. When the recognition system was based on the two retinas, the false non-match rate achieved was 2.66 × 10-7, or approximately one false non-match in 3.7 million matching attempts. © 2011 American Society of Agricultural and Biological Engineers.