Background: There has been limited acceptance of shoulder ultrasonography by orthopaedic surgeons in the United States. The purpose of this retrospective study was to determine the diagnostic performance of high- resolution ultrasonography compared with arthroscopic examination for the detection and characterization of rotator cuff tears. Methods: One hundred consecutive shoulders in ninety-eight patients with shoulder pain who had undergone preoperative ultrasonography and subsequent arthroscopy were identified. The arthroscopic diagnosis was a full-thickness rotator cuff tear in sixty-five shoulders, a partial-thickness tear in fifteen, rotator cuff tendinitis in twelve, frozen shoulder in four, arthrosis of the acromioclavicular joint in two, and a superior labral tear and calcific bursitis in one shoulder each. All ultrasonographic reports were reviewed for the presence or absence of a rotator cuff tear and a biceps tendon rupture or dislocation. All arthroscopic examinations were performed according to a standardized operative procedure. The size and extent of the tear and the status of the biceps tendon were recorded for all shoulders. The findings on ultrasonography and arthroscopy then were compared for each parameter. Results: Ultrasonography correctly identified all sixty-five full-thickness rotator cuff tears (a sensitivity of 100 percent). There were seventeen true- negative and three false-positive ultrasonograms (a specificity of 85 percent). The overall accuracy was 96 percent. The size of the tear on transverse measurement was correctly predicted in 86 percent of the shoulders with a full-thickness tear. Ultrasonography detected a tear in ten of fifteen shoulders with a partial-thickness tear that was diagnosed on arthroscopy. Five of six dislocations and seven of eleven ruptures of the biceps tendon were identified correctly. Conclusions: Ultrasonography was highly accurate for detecting full-thickness rotator cuff tears, characterizing their extent, and visualizing dislocations of the biceps tendon. It was less sensitive for detecting partial-thickness rotator cuff tears and ruptures of the biceps tendon.