Between 1976 and 2004, 38 revision arthroplasties (35 patients) were performed for aseptic loosening of the humeral component. The mean interval from primary arthroplasty to revision was 7.1 years (0.4 to 16.6). A total of 35 shoulders (32 patients) were available for review at a mean follow-up of seven years (2 to 19.3). Pre-operatively, 34 patients (97%) had moderate or severe pain; at final follow-up, 29 (83%) had no or only mild pain (p < 0.0001). The mean active abduction improved from 88 degrees to 107 degrees (p < 0.01); and the mean external rotation from 37 degrees to 46 degrees (p = 0.27). Excellent or satisfactory results were achieved in 25 patients (71%) according to the modified Neer rating system. Humeral components were cemented in 29, with ingrowth implants used in nine cases. There were 19 of standard length and 17 were longer (two were custom replacements and are not included). Bone grafting was required for defects in 11 humeri. Only two glenoid components were left unrevised. Intra-operative complications included cement extrusion in eight cases, fracture of the shaft of the humerus is two and of the tuberosity in four. There were four re-operations, one for recurrent humeral loosening, with 89% survival free of re-operations at ten years. Revision surgery for aseptic loosening of the humeral component provides reliable pain relief and modest improvement of movement, although there is a substantial risk of intra-operative complications. Revision to a total shoulder replacement gives better results than to a hemiarthroplasty.