OBJECTIVE. The purpose of our study was to determine the interobserver variability in the sonographic evaluation of the rotator cuff. SUBJECTS AND METHODS. Two radiologists independently scanned 61 patients with shoulder pain. Each radiologist categorized the rotator cuff as normal, partially torn, or fully torn. When a tear was present, the tendons involved were specified. All diagnoses were made prospectively without knowledge of the findings of the other radiologist. RESULTS. The radiologists were in full agreement in the categorization of 92% (56/61) of the patients. In four of the five discrepant cases, the disagreement was whether there was a full-thickness or a partial-thickness tear. The radiologists were in agreement concerning which tendons were involved in 80% (41/51) of the patients in whom a tear was detected by both observers. In all 10 discrepant cases, the disagreement was whether a tear involved both the supraspinatus and infraspinatus tendons or was isolated to one or the other of these tendons. CONCLUSION. The level of interobserver variability in the sonographic detection and characterization of rotator cuff tears is low.