Our purpose was to compare the accuracy of ultrasonography (US) and magnetic resonance (MR) in the diagnosis of rotator cuff injuries (focusing on supraspinatus tears) using arthroscopy or open surgery findings. Using US and MR, we studied all painful shoulders seen during 1998 in our orthopaedic unit. Diagnosis was confirmed in 61 cases. The diagnosis of full thickness rotator cuff tear was highly specific on both imaging techniques (100% for US and 97.1% for MR) but was not as sensitive on US (57.7%) as on MR (80.8%). The diagnosis of partial tear was acceptably specific (67.9% for US and 75.5% for MR) but was not sensitive (overall, 12.5% for US vs 50% for MR). The specificity of a US diagnosis of a tear allows us to recommend surgery, thus making it a good imaging technique for the initial study of painful shoulders. The low sensitivity makes it necessary to undertake an additional MR study. If the MR also discovers no pathology, the sensitivity based on the US and MR images (92.7%) can be sufficient to avoid a false-negative diagnosis.