Thirty-seven elbows in thirty-six patients who had rheumatoid arthritis had a total elbow arthroplasty with insertion of a non-constrained surface-replacement prosthesis. The patients were followed for an average of nine years and six months, the longest follow-up being seventeen years. A good result was seen in twenty-nine elbows; a fair result, in one; and a poor result, in seven. The reasons for the poor results were gross posterior displacement of the humeral component in five elbows, persistent subluxation with pain in one, and recurrent ankylosis in one. Of the five elbows that had gross posterior displacement, four had a revision operation with a new humeral component, and a satisfactory result was eventually achieved. Radiographic examination revealed various degrees of proximal subsidence of the humeral component in 70 per cent of the elbows. However, in most of the elbows the subsidence was not progressive and was compatible with a good clinical result. In contrast, the rate of loosening of the ulnar component was low; loosening was seen in only 5 per cent of the elbows. Seventeen elbows were followed for ten years or more, and comparison of the clinical results in the intermediate period with those at the most recent review revealed that the results improved with time. Because of the number of elbows in which subsidence of the humeral component developed, we now use a humeral component with an intramedullary stem, and were no longer recommend the use of our Type-1 and Type-2 prostheses.