Six thousand four hundred eighty-nine knee replacements were done in 6120 patients at the authors' institution between 1993 and 1999. Operations were done in a theater with vertical laminar flow and with the surgical team using body exhaust suits. Of these knee replacements, 116 knees became infected and 113 were available for followup. One hundred of the infections occurred in patients undergoing primary knee replacement, whereas the remaining infections occurred in patients undergoing revision knee replacement. Ninety-seven of these knees (86%) had deep periprosthetic infections and the remaining 16 knees had superficial wound infections. One third of the deep infections occurred within the first 3 months after surgery and the remaining ⅔ occurred after 3 months. The overall early deep infection rate for patients undergoing a primary knee replacement was 0.39%, whereas the rate for patients undergoing a revision knee replacement was 0.97%. A cohort of noninfected knee replacements from patients matched for gender, age, and month of surgery was used as a control group. Those comorbidities that were statistically significant in increasing the risk of infection were prior open surgical procedures, immunosuppressive therapy, poor nutrition, hypokalemia, diabetes mellitus, obesity, and a history of smoking. Patients undergoing revision procedures had a statistically higher risk of infection than did patients undergoing primary surgeries. If the surgery took longer than 2.5 hours, the risk of infection was increased significantly. There was no change in the infection rate when the perioperative antibiotic prophylaxis was decreased from 48 to 24 hours after surgery. The predominant infectious organisms were gram-positive (Staphylococcus aureus, Staphylococcus epidermidis, and Streptococcus Group B). Twenty percent of the knees that were infected clinically had no organisms that could be identified. In each case, the patient had been treated empirically at another institution with antibiotics before a culture of the joint was obtained.