The true incidence of bedside transfusion errors, i.e. those happening when blood products have left the blood bank, is underestimated because published figures rely on reporting of clinically relevant events or on indirect methods. The SAnGUIS project assessing blood practice in a prospective and randomized fashion for 6 elective surgical procedures gave the opportunity to trace all transfused units and to identify steps at risk during blood delivery in surgery. We considered transfusion of a wrong unit as a major error and poor execution or documentation as a recording error. Over 15 months, 808 patients out of 1,448 were transfused with 3,485 units. A total of 165 errors were found after blood products had left the blood banks. Seven were misidentifications (0.74% of patients, 0.2% of units). Eight other major errors occurred in 4 (0.5%) patients. Major errors occurred during nonemergency situations, in wards or intensive care units. The remaining ('recording') 150 errors consisted of misrecordings (61), mislabellings (6), or failures to document transfusions in the medical records (83). All errors were uneventful except one misidentification which induced a transient, yet unreported, reaction. The 'descending' inquiry method used for this study showed that most errors pass unnoticed and are therefore not reported. Measurement of error rates may constitute an important quality indicator. Retrospective information of this survey to the concerned staff people provided an impetus to take adequate measures to reduce these bedside errors.