This study is both a retrospective and prospective evaluation of the clinical usefulness of shoulder sonography. Ninety-eight patients suspected of having rotator cuff tears underwent sonography of both shoulders. Sixty-two patients underwent double-contrast arthrography performed on the same day as sonography, and 38 patients underwent surgery after sonography. A comparison of the results from ultrasound and arthrography, using published diagnostic criteria, demonstrated a sensitivity of 75% and a specificity of 43% for detection of a rotator cuff tear. In this study, use of more restricted criteria, a subset of the published criteria, yielded a sensitivity of 68% and a specificity of 90%. A comparison of sonography with surgery, using this study's criteria, demonstrated a sensitivity of 57% and a specificity of 76%. This report shows that shoulder sonography is less reliable than previously reported and appears to have a very limited role in the evaluation of rotator cuff injuries.