The accuracy of ultrasonography and arthrography in the prediction of lesions of the rotator cuff was evaluated and compared with the operative findings in a retrospective study of forty-nine patients (forty-nine shoulders). In the detection of full-thickness (stage-III) tears of the rotator cuff, ultrasonography had an over-all sensitivity of 74 per cent, a specificity of 95 per cent, an accuracy of 84 per cent, a predictive value of a positive test of 95 per cent, and a predictive value of a negative test of 75 per cent. In contrast, arthrography demonstrated a sensitivity of 93 per cent, a specificity of 95 per cent, an accuracy of 94 per cent, a predictive value of a positive test of 96 per cent, and a predictive value of a negative test of 91 per cent. Ultrasonography could not reliably differentiate between partial (stage-II) and full-thickness (stage-III) tears. Lesions of the long head of the biceps brachii tendon were recorded by ultrasonography for twenty-two patients (45 per cent) and by arthrography for eleven patients (22 per cent). The ultrasonographic finding of fluid in the sheath of this tendon is not a reliable sign of a lesion in the tendon.