Cardiopulmonary bypass (CPB) involves the use of either an occlusive roller pump or centrifugal pump. Damage to blood elements, including haemolysis, may arise from occlusion when using a roller pump; the appropriate degree of occlusion has not yet been determined scientifically. Centrifugal and nonocclusive roller pumps are reputed to reduce haemolysis. The objective of this study was to compare haemolysis caused by a standard roller pump with a dynamically set nonocclusive roller pump and with a centrifugal pump. We prospectively randomized 60 patients undergoing routine coronary artery surgery into three groups: standard roller pump (STD, n = 20), dynamically set roller pump (DYN, n = 20), or centrifugal pump (CEN, n = 20). The level of plasma free haemoglobin (FHb) was measured preoperatively, and the rate of formation of FHb (in mg/dl/min) was determined at the end of the ischaemic phase and at the end of CPB. Cardiotomy suction blood was isolated for the ischaemic phase and returned before the end of CPB. It was found that there were no differences between the groups in demographic or operative variables. The rate of formation of FHb at the end of the ischaemic phase was similar for all groups (STD 0.108 +/- 0.10, DYN 0.117 +/- 0.08, CEN 0.129 +/- 0.07). At the end of CPB, after return of the cardiotomy suction blood, there was a significant (< 0.001) increase in the rate of formation of FHb in all groups. The increase was similar for each of the groups (STD 0.424 +/- 0.17, DYN 0.481 +/- 0.20, CEN 0.471 +/- 0.18). We conclude that the rates of haemolysis are similar for each of the pump types, and no benefit is conferred by the use of either a dynamically set roller pump or a centrifugal pump compared with the standard roller pump. The return of the cardiotomy suction blood to the circulation is the principal source of plasma free haemoglobin.