To prevent haemodynamic stroke during cardiovascular surgery in patients with carotid stenosis, we routinely evaluated magnetic resonance angiography and selectively evaluated brain perfusion single-photon emission computed tomography with acetazolamide challenge. Off-pump surgery was preferred when cerebral blood flow reserve was impaired. This strategy's usefulness was investigated. Among the 1059 consecutive patients who underwent preoperative carotid screening by magnetic resonance angiography, 84 (7.9%) patients had >50% stenosis; 45 of them underwent brain single-photon emission computed tomography. The severity of cerebral blood flow compromise was estimated by the proportion of Stage 2 area in the affected territory, in which both resting blood flow (<32 ml/min) and flow reserve (<10%) were reduced. Perioperative stroke occurred in 1.7% overall (18/1059), in 6% (5/84) of those with carotid stenosis and in 1.3% (13/975) of those without stenosis (P = 0.010). On subgroup analysis, carotid stenosis was associated with an increased risk of stroke in the on-pump surgery group [n = 949, 5/59 (9%) with stenosis vs 11/890 (1.1%) without stenosis, P = 0.002], while it was not in the off-pump group [n = 110, 0/25 (0%) with stenosis vs 2/85 (2%) without stenosis, P = 0.59]. With respect to the role of acetazolamide single-photon emission computed tomography, 2 of the 4 patients with Stage 2 area >10% undergoing on-pump surgery without preceding carotid revascularization developed stroke, while none of the 21 patients with Stage 2 area <10% undergoing on-pump surgery developed stroke (P = 0.020). Carotid stenosis is a risk factor for perioperative stroke in on-pump surgery. Patients with large Stage 2 area (>10%) are at increased risk of perioperative stroke when on-pump surgery is performed.