Twenty-one patients who had had pain in the shoulder for more than three months were evaluated with ultrasonography and magnetic resonance imaging followed by computerized tomographic arthrography. The results of the imaging studies were then compared with the operative findings. Magnetic resonance imaging was found to be the most useful modality for establishment of the etiology of pain in the shoulder due to disease of the rotator cuff, instability associated with abnormality of the glenoid labrum, subacromial impingement, stenosis of the coracoacromial arch, and osteoarthrosis of either the glenohumeral or the acromioclavicular joint. The accuracy of magnetic resonance imaging was found to depend on both the operator and the technique and was decreased in extremely obese patients, due to difficulties in positioning, and in patients who had had a previous operation. Magnetic resonance imaging was more accurate than either computerized tomographic arthrography or ultrasonography in identifying partial-thickness tears (intrasubstance changes in the rotator cuff). Magnetic resonance imaging provided the same level of accuracy as computerized tomographic arthrography in the detection of abnormalities of the glenoid labrum.