Thirty-one symptomatic patients were studied with MR imaging to evaluate the sensitivity and specificity of shoulder MR in the diagnosis of rotator cuff tears. Correlative studies included arthroscopy in 19 patients and arthrography in 12 patients. Images were obtained on either a 0.5- or 1.5-T Philips superconducting magnet using spin-echo pulse sequences (650-850/30 [TR, TE], 2000/30, 100) with 5-mm slices oriented in an oblique coronal plane perpendicular to the glenohumeral joint. The MR studies were initially interpreted without knowledge of the results of other diagnostic procedures. The MR diagnosis of cuff tear was made when irregularity, discontinuity, and increased signal were identified in the rotator cuff. MR images showed tears in 10 patients (32%) and were negative for tear in 21 patients (68%). MR correlated with arthroscopy and arthrography in 17 of 18 normal patients, in eight of 10 patients with complete tears, and in one of three patients with partial tears. For complete rotator cuff tears, the sensitivity, specificity, and accuracy were 80%, 94%, and 89%, respectively. For all tears (partial and complete), the sensitivity, specificity, and accuracy were 69%, 94%, and 84%, respectively. These data suggest that MR imaging is an accurate procedure for the diagnosis of complete rotator cuff tears. The number of partial tears (three) in this series is too small to evaluate the value of MR imaging in the diagnosis of partial tears.