The initiation and progression of atherosclerosis, which is the main cause of cardiovascular diseases, correlate with local haemodynamic factors such as wall shear stress (WSS). Numerical simulations such as computational fluid dynamics (CFD) based on medical imaging have been employed to analyse blood flow in different arteries with and without luminal stenosis. Patient-specific CFD models, however, have assumptions on blood rheology. The differences in the calculated haemodynamic factors between different rheological models have not been fully evaluated. In this study, carotid magnetic resonance imaging (MRI) was performed on six patients with different degrees of carotid stenosis and two healthy volunteers. Using the 3D reconstructed carotid geometries and the patient-specific boundary conditions, CFD simulations were performed by applying a Newtonian and four non-Newtonian models (Carreau, Cross, Quemada and Power-law). WSS descriptors and pressure gradient were analysed and compared between the models. The differences in the maximum and the average oscillatory shear index between the Newtonian and the non-Newtonian models were lower than 12.7% and 12%, respectively. The differences in pressure gradient were also within 15%. The differences in the mean time-averaged WSS (TAWSS) between the Newtonian and Cross, Carreau and Power-law models were lower than 6%. In contrast, a higher difference (26%) was found in Quemada. For the low TAWSS, the differences from the Newtonian to the non-Newtonian models were much larger, in the range of 0.4–31% for Carreau, 3–22% for Cross, 5–51% for Quemada and 10–41% for Power-law. The study suggests that the assumption of a Newtonian model is reasonable when the overall flow pattern or the mean values of the WSS descriptors are investigated. However, the non-Newtonian model is necessary when the low TAWSS region is the focus, especially for arteries with severe stenosis.