To date, no published data exist confirming the reliability of the Neer, Levigne, or Miyoshi classifications for glenohumeral wear patterns in rheumatoid arthritis. We examined the intraobserver and interobserver agreement of these systems to determine which is most reliable. Four orthopedic surgeons with fellowship training in upper extremity disorders and 4 orthopedic trainees examined 53 radiographs of rheumatoid shoulders and classified them according to the 3 radiographic schemes. After a minimum 2-week interval, each observer repeated the classification process. The Levigne classification demonstrated good intraobserver (κ=0.80) and interobserver agreement (κ=0.64); the Neer classification, good intraobserver (κ=0.61) and moderate interobserver agreement (κ=0.50); and the Miyoshi classification, good intraobserver (κ=0.66) and fair interobserver agreement (κ=0.37). There were no differences in intraobserver reliability between trainees and fellowship-trained physicians. Although all systems demonstrated reasonable overall reliability, these data suggest that the Levigne system may be most reliable for classification of rheumatoid arthritis affecting the glenohumeral joint. Level of training did not impact the ability of observers to reliably judge rheumatoid wear patterns with these classification systems.