The blood flow characteristics found in our larger vessels are unsteady, particularly around the heart valves and bifurcations. In the case of stenosis, or narrowing of the vessels, the flow may transition to turbulence. To understand the dynamics of the forces acting on the blood components and the vessel wall, simulations using computational fluid dynamics (CFD) are commonly applied. The severity of the stenosis can be determined by accurately assessing the fluid flow, which can also serve as a risk indicator for potential thromboembolic events. Motivated by the vessel’s geometry being a factor that highly influences the flow characteristics, we investigate here the impact of changes in geometry on turbulence using multi-fidelity models, which are based on Gaussian processes. The objective is to develop a multi-fidelity model to construct a high-fidelity estimate by combining numerical simulations from spectral element-based direct numerical simulations (DNS) and finite volume-based Reynolds-Averaged Navier-Stokes (RANS) simulations. Specifically, a co-kriging-based model with Gaussian process is used to combine various levels of fidelity (RANS, DNS). To vary the blood vessel geometry, the stenosis’s severity and eccentricity are considered uncertain input parameters. A multi-fidelity model is then used to predict the consequences of said uncertainties on the mean pressure drop across the vessel and the wall shear stress, the quantities of interest directly linked to the biological activity of the vessel. Using data of different accuracy, the multi-fidelity technique allows us to optimize the accuracy and cost of predictions.