Whether third‐generation hydroxyethyl starch solutions provoke kidney injury or haemostatic abnormalities in patients having cardiac surgery remains unclear. We tested the hypotheses that intra‐operative administration of a third‐generation starch does not worsen postoperative kidney function or haemostasis in cardiac surgical patients compared with human albumin 5%. This triple‐blind, non‐inferiority, clinical trial randomly allocated patients aged 40–85 who underwent elective aortic valve replacement, with or without coronary artery bypass grafting, to plasma volume replacement with 6% starch 130/0.4 vs. 5% human albumin. Our primary outcome was postoperative urinary neutrophil gelatinase‐associated lipocalin concentrations, a sensitive and early marker of postoperative kidney injury. Secondarily, we evaluated urinary interleukin‐18; acute kidney injury using creatinine RIFLE criteria, coagulation measures, platelet count and function. Non‐inferiority (delta 15%) was assessed with correction for multiple comparisons. We enrolled 141 patients (69 starch, 72 albumin) as planned. Results of the primary analysis demonstrated that postoperative urine neutrophil gelatinase‐associated lipocalin (median (IQR [range])) was slightly lower with hydroxyethyl starch (5 (1–68 [0–996]) ng.ml⁻¹) vs. albumin (5 (2–74 [0–1604]) ng.ml⁻¹), although not non‐inferior [ratio of geometric means (95%CI) 0.91 (0.57, 1.44); p = 0.15] due to higher than expected variability. Urine interleukin‐18 concentrations were reduced, but interleukin‐18 and kidney injury were again not non‐inferior. Of 11 individual coagulation measures, platelet count and function, nine were non‐inferior to albumin. Two remaining measures, thromboelastographic R value and arachidonic acid‐induced platelet aggregation, were clinically similar but with wide confidence intervals. Starch administration during cardiac surgery produced similar observed effects on postoperative kidney function, coagulation, platelet count and platelet function compared with albumin, though greater than expected variability and wide confidence intervals precluded the conclusion of non‐inferiority. Long‐term mortality and kidney function appeared similar between starch and albumin.