Few randomized controlled studies, the only trial design where causality can be established between an intervention and the benefits or harms thereof, have been published on the benefits and risks of a restrictive vs a liberal transfusion strategy. We review the 19 controlled studies on erythrocyte transfusion thresholds published since the eighties. These studies suggest that, overall, morbidity (including cardiac morbidity) and mortality, along with hemodynamic, respiratory and oxygen transport variables, are similar when a restrictive transfusion strategy (transfusion threshold between 7 and 8 g/dL) or a liberal strategy (transfusion threshold of 10 g/dL) are used. In fact, a restrictive strategy can even be associated with a number of benefits. The relevance of a higher transfusion threshold in view of avoiding morbidity in patients presenting a cardiovascular risk is unlikely, at least uncertain. Finally, anaemia has little or no impact on functional recovery and on quality of life, whether in the immediate or late postoperative period. It is clear that a restrictive strategy is associated with a reduced exposure to red cell transfusions, allowing a reduction in transfusion-related adverse events. Thus, all red cell transfusions must be tailored to the patient's needs, at the time the need prevails. In conclusion, most recommendations on transfusion practice are limited by the lack of evidence-based data and reveal our ignorance on the topic. High quality clinical trials in different patient populations must become available in order to determine optimal transfusion practices. Since then, a restrictive strategy aiming for a moderately anaemic threshold (7-8 g/dL) is appropriate under most circumstances.