To determine the effect of transfusion on the incidence of postoperative infection, a retrospective cohort study of 196 patients who underwent surgery for gastric carcinoma in the period from 1985 through 1989 was carried out. Seventy-one patients (36.2%) developed postoperative septic complications; they had received an average of 4.2 blood units, as compared with 2.7 units received by patients not affected (p less than 0.0053). The hypothesis of dose-response relationship is supported by the Mantel-Haenszel test, as applied to the overall results (p less than 0.01) and the results grouped by duration of operation (p less than 0.02). Furthermore, logistic regression analysis shows transfusion to be an independent risk factor in the incidence of infection (p less than 0.01), as are antibiotic prophylaxis (p less than 0.015), urinary tract catheterization (p less than 0.002), and the duration of surgery (p less than 0.027). This significance is attained after adjustment for age, gender, period of evolution of symptoms; preoperative infection(s), number of white cells, hemoglobin level and total proteins on diagnosis, location of tumor, tumor, nodes, and metastasis staging, operative technique, drainage of the area of operation, enteral nutrition, and the histologic studies and macroscopic appearance of the tumor. This study is further evidence that transfusion may cause an increased incidence of postoperative infection.