Total knee arthroplasty is a predictable operation. Unfortunately, there is a subset of patients who do not do well and require revision surgery within the first 5 years. The purpose of the current study was to analyze the mechanisms of failure in patients who had revision surgery within 5 years of their index arthroplasty. Between 1986 and 1999, 440 patients with total knee arthroplasties were referred for revision surgery. An analysis of patients in whom the arthroplasties failed within 5 years of the index arthroplasty and the reasons for early failure were documented. Of the 440 patients who had revision surgery, 279 (63%) had revision surgery within 5 years of their index arthroplasty: 105 of the 279 patients with early failures (38%) had revision surgery because of infection; 74 (27%) had revision surgery because of instability; 37 (13%) had revision surgery because of failure of ingrowth of a porous-coated implant; 22 (8%) had revision surgery because of patellofemoral problems; and 21 (7%) had revision surgery because of wear or osteolysis. Only eight of the 279 patients with early failures (3%) had revision surgery because of aseptic loosening of a cemented implant. The remaining 12 patients had revision surgery because of miscellaneous problems. Host factors may prevent infection from ever being eradicated totally. The two other major patterns of failure in this series were failure of cementless fixation and instability. If all of the arthroplasties in the patients in this early failure group would have been cemented routinely and balanced carefully, the total number of early revisions would have decreased by approximately 40%, and the overall failures would have been reduced by 25%.