Seventy-eight Neer hemiarthroplasties and thirty-six Neer total shoulder arthroplasties were performed at our institution, between January 1, 1976, and December 31, 1985, in ninety-eight patients who were fifty years old or less. Two patients (two shoulders) died, and four patients (four shoulders) were lost to follow-up. The remaining seventy-four hemiarthroplasties (95 per cent) in sixty-four patients and thirty-four total shoulder arthroplasties (94 per cent) in thirty-one patients were included in the clinical analysis as the preoperative and operative records were complete and the patients had been followed for at least five years (mean, 12.3 years) or until revision. All 114 shoulders were included in the survivorship analysis.Both total shoulder arthroplasty and hemiarthroplasty resulted in significant long-term relief of pain (p < 0.0001) as well as improvement in active abduction (p < 0.0001) and external rotation (p < 0.0001). However, with the numbers available, we could not detect a significant difference between the two procedures with respect to these variables.A complete set of radiographs was available for sixty-eight (92 per cent) of the seventy-four shoulders that had a hemiarthroplasty and for thirty-two (94 per cent) of the thirty-four shoulders that had a total shoulder arthroplasty. A radiolucent line around the humeral component was noted after sixteen (24 per cent) of the hemiarthroplasties and after seventeen (53 per cent) of the total shoulder arthroplasties. A radiolucent line around the glenoid component was seen after nineteen (59 per cent) of the total shoulder arthroplasties. Erosion of the glenoid was found after forty-six (68 per cent) of the hemiarthroplasties.The results were graded according to a modification of the system of Neer et al. and of Cofield. Fifteen hemiarthroplasties led to an excellent result; twenty-four, a satisfactory result; and thirty-five, an unsatisfactory or unsuccessful result. Four total shoulder arthroplasties were followed by an excellent result; thirteen, a satisfactory result; and seventeen, an unsatisfactory or unsuccessful result. The estimated survival of the hemiarthroplasty prostheses (with 95 per cent confidence intervals) was 92 per cent (86 to 98 per cent) at five years, 83 per cent (75 to 93 per cent) at ten years, and 73 per cent (59 to 88 per cent) at fifteen years. Analysis of the results in association with the two major diagnoses revealed that the risk of revision was higher for the thirty shoulders that had the hemiarthroplasty for the treatment of the sequelae of trauma than for the twenty-eight that had the procedure for the treatment of rheumatoid arthritis (p = 0.017). The estimated survival of the total shoulder prostheses (with 95 per cent confidence intervals) was 97 per cent (92 to 100 per cent) at five years, 97 per cent (91 to 100 per cent) at ten years, and 84 per cent (70 to 100 per cent) at fifteen years. The risk of revision was higher for the seven shoulders that had had a tear of the rotator cuff at the time of the operation than for the twenty-seven that had not had one (p = 0.029).The data from the present study indicate that a shoulder arthroplasty provides marked long-term relief of pain and improvement in motion; however, nearly half of all young patients who have a shoulder arthroplasty have an unsatisfactory result according to a rating system. Care should be exercised when either a hemiarthroplasty or a total shoulder arthroplasty is offered to patients who are fifty years old or less.