In many patients, a 'type and screen' procedure is routinely performed before surgery. However, most patients are not transfused after all. Can we predict, which surgical patients will and will not be transfused, to reduce the number of these investigations? We studied 1482 consecutive surgical patients with intermediate risk for transfusion. Multivariate logistic regression modelling and the area under the Receiver Operating Characteristic curve (ROC area) were used to quantify how well age, gender, surgical procedure, emergency or elective surgery and anaesthetic technique predicted transfusion, and whether the preoperative haemoglobin concentration had added predictive value. Gender, age > or =70 yr, and type of surgery were independent predictors of transfusion, with a ROC area of 0.75 (95% CI: 0.72-0.79). Validating this model with an easily used prediction rule in a second patient population yielded a ROC area of 0.70 (95% CI: 0.63-0.77). With this rule type and screen could correctly be withheld in 35% of these patients. In the remaining 65% of the patients, a further reduction in type and screen investigations of 15% could be achieved using the preoperative haemoglobin concentration. Using a simple prediction rule, preoperative type and screen investigations in patients who have to undergo surgery procedures with intermediate transfusion risk can be avoided in about 50%. This may reduce patient burden and hospital costs (on average: 3 million US$ per 100 000 procedures).