OBJECTIVE: To determine the interobserver and intraobserver reliability of the interpretation of MRIs for supraspinatus tendinosis. METHODS: In the interobserver trial, the MRIs of 52 athletes' shoulders were observed by 3 observers on one occasion within a 2-month period. All 52 images were read by the most experienced musculoskeletal radiologist on 3 different occasions on separate days without access to the previous readings for the intraobserver trial. Supraspinatus tendinosis was graded using a modified 4-point scale from grades 0 to grade 3. RESULTS: The grading of MRI-determined supraspinatus tendinosis was reliable, having an intraclass correlation (ICC) of 0.85 when assessed by the single well-trained observer. Interobserver reliability was only fair to good (ICC = 0.55). CONCLUSIONS: Supraspinatus tendinosis can be accurately identified on MRI with little variation by a single well-trained observer. Interobserver reliability was only fair to good. Our data indicated that the reliability of the assessment was much greater in more experienced radiologists than in those with less experience.