We have undertaken a prospective clinical and radiological analysis of 124 shoulder arthroplasties (113 patients) carried out for osteoarthritis. The clinical results showed improvement in the absolute Constant score and the American Shoulder and Elbow Surgeons score of 22 and 43, respectively. Both were statistically significant (p < 0.001). There was no significant difference in the scores after hemiarthroplasty and total arthroplasty in those patients with an intact rotator cuff. When revision was used as the end-point for survival at ten years, survival of 86%, or 90% if glenoid components made of Hylamer sterilised in air were omitted, was obtained in primary osteoarthritis. The most common cause for revision in the hemiarthroplasty group was glenoid pain at a mean of 1.5 years; in the total arthroplasty group it was loosening of the glenoid at a mean of 4.5 years. Analysis of pre-operative factors showed that the risk of gross loosening of the glenoid increased threefold when there was evidence of erosion of the glenoid at operation. Shoulder arthroplasty should not be delayed once symptomatic osteoarthritis has been established and should be undertaken before failure of the cuff or erosion of the glenoid are present.