We prospectively studied 123 patients with colorectal cancer in order to identify determinants of perioperative blood transfusions and unnecessary transfusions. Transfusions were considered unnecessary if the preoperative hematocrit reading exceeded 36% or the discharge hematocrit level exceeded 33%. Age, sex, admission hematocrit reading, operative procedure, specimen length, duration of surgery, estimated blood loss, tumor size, tumor differentiation, nodal status, Dukes' staging, and attending surgeon were evaluated in relation to perioperative blood transfusion using stepwise logistic regression. Fifty-one (41%) of the 123 patients in the study were transfused and 35 patients (28%) received at least 1 unnecessary unit of blood. Advanced age, low admission hematocrit reading, high estimated operative blood loss, prolonged procedures, and lengthy specimens were significantly related to the administration of blood. Excessive intraoperative transfusions and the practice of administering blood in pairs of units accounted for the unnecessary transfusions. This study indicates that factors that influence clinical judgment and thereby the decision to transfuse a patient are not accurate indicators of when blood should be given. Greater use of pre-transfusion hematocrit testing, especially in the operating room and before transfusing a second unit of blood, will dramatically reduce blood usage in elective colorectal cancer surgery.