Preoperative autologous blood donation is employed with increasing frequency, particularly in patients undergoing elective orthopedic procedures. While autologous transfusion decreases the incidence of postoperative infections and other complications, the cost-effectiveness of this therapy has not been fully investigated. We constructed a decision analytic model to study the cost-effectiveness of preoperative autologous blood donation of packed red blood cells compared with allogeneic packed red blood cells in primary hip arthroplasty. We used data from 73 patients presenting at our blood center with a prescription for 2 U of autologous red blood cells prior to hip arthroplasty to establish probabilities for the number of units that would be donated. Patients were able to donate an average of 1.9 U (range, 0 to 2 U) of autologous blood. We also reviewed the charts of 56 patients who underwent primary hip arthroplasty to model the number of units given during hospitalization (1.5 U given; range, 0 to 5 U). We applied the model to a 65-year-old patient undergoing primary hip arthroplasty. Estimates for the incidence of posttransfusion hepatitis, chronic active hepatitis, human immunodeficiency virus infection, postoperative bacterial infection, and fatal hemolytic transfusion reaction were derived from the literature. Patient utility was measured in life-years. Costs included the cost of preoperative autologous blood donation, blood administration, and medical care costs associated with the complications of transfusion. Costs were derived from local data and the literature. Future earnings lost were not modeled. In the baseline analysis, autologous transfusion results in a net cost savings compared with allogeneic blood over a wide range of complication rates, patient ages, and transfusion requirements. The dominant factor in the analysis is the effect of postoperative bacterial infection on length of hospital stay and the resultant increase in costs. The effect of viral infections on the results of the analysis is minimal.