In 28 patients the preoperative MR results and arthroscopic examination findings, in which both sides of the rotator cuff were systematically examined, were compared. The interpretive MR criteria were both sensitive and specific for detection of full thickness tears: 100% (5 of 5) and 100% (23 of 23), respectively. Criteria were insensitive in 71% (10 of 14) but specific in 93% (13 of 14) for detection of all tears; 4 of 9 partial thickness tears were not detected of MR. The potential significance of these tears and the use of fat suppression MR sequences to improve detection and depiction of rotator cuff abnormalities are discussed.