Magnetic resonance imaging (MRI) is used increasingly for evaluating the rotator cuff. This study of 39 shoulders (38 patients) compared the accuracy of MRI interpretation of rotator cuff integrity by a group of community hospital radiologists (clinical community scenario, CCS) with that of a musculoskeletal radiologist (experienced specialist scenario, ESS), relative to arthroscopy. For the CCS subgroup, the sensitivity, specificity, positive predictive value (PV), negative PV, and accuracy for partial tears were: 0%, 68%, 0%, 82%, and 59%, respectively; for complete tears: 56%, 73%, 36%, 86%, and 69%, respectively; and for all tears combined: 85%, 52%, 50%, 87%, and 64%, respectively. For the ESS subgroup, the respective values for partial tears were: 20%, 88%, 20%, 88%, and 79%, respectively; for complete tears: 78%, 83%, 58%, 92%, and 82%, respectively; and for all tears: 71%, 71%, 59%, 81%, and 71%, respectively. We concluded that MRI assessment of the rotator cuff was not accurate relative to arthroscopy. MRI was most helpful if the result was negative, and MRI diagnosis of partial tear was of little value. Considering the high cost of shoulder MRI, this study has significant implications for the evaluation of patients with possible rotator cuff pathology.