Thirty-nine patients (forty shoulders) with primary osteoarthritis consented to be randomized to receive either a cemented all-polyethylene glenoid component or a cementless metal-backed component at the time of total shoulder arthroplasty. Their mean age was 69 years. Preoperative and postoperative evaluations were completed at 3, 6, 12, 24, and 36 months by history, physical examination, radiographs, and Constant scoring system. The presence of periprosthetic radiolucent lines was significantly greater with polyethylene than with metal-backed glenoids (85% vs 25%, P <.01). Of 20 radiolucent lines, 12 (60%) around polyethylene glenoids were present on immediate postoperative radiographs and 25% were progressive. No significant correlation was found between the presence of radiolucent lines around polyethylene glenoids and functional results (P =.3). By contrast, periprosthetic radiolucent lines around metal-backed glenoids were rare but progressive when present. The incidence of loosening of metal-backed implants (4 cases, 20%) was significantly higher than that observed with polyethylene glenoids (0%, P <.001) and was associated with component shift and severe osteolysis. Metal-backed glenoid loosening significantly correlated with deteriorating functional results and increasing pain (P <.05). Revision surgery was required for 4 patients in the metal-backed group (P =.02), for a subscapularis tear (1 case) and metal-backed glenoid component loosening (3 cases). Computed tomography scan analysis and revision surgery revealed that preoperative posterior humeral subluxation may recur with time despite glenoid reorientation and may cause asymmetric accelerated polyethylene wear, resulting in metal-on-metal contact and severe osteolysis. Reimplantation of a stable cemented glenoid component was possible in 1 case, whereas the cavitary defect was packed with cancellous bone in the 2 other cases. At a minimum of 3 years' follow-up, the results of this study clearly show that (1) the survival rate of cementless, metal-backed glenoid components is inferior to cemented all-polyethylene components and (2) the incidence of radiolucency at the glenoid-cement interface with all-polyethylene components is high and remains a concern. The high rate of loosening, because of the absence of ingrowth and/or the accelerated polyethylene wear, has led us to abandon the use of metal-backed glenoids. Efforts must continue to improve glenoid component design and fixation.