This study proposes to quantify the uncertainty in a CPU time costly computational fluid dynamics (CFD) model used to evaluate the local temperature field in the situation of blocked fuel assembly in a pressurized water reactor (PWR) transfer tube. Several uncertain parameters are identified and a first uncertainty propagation study is conducted on a low-fidelity (poorly refined) mesh for CPU cost issues. Then, using the concept of “support points,” an algorithm is employed to reduce the size of the initial design of experiments. A high-fidelity model (finer mesh, more CPU time expensive) is then run on this small-size design of experiments. A metamodel was finally built on those high-fidelity results to propagate uncertainties and finely analyze the results. The successful results that are obtained show that metamodeling has the potential to overcome the issue of costly CPU time CFD models in the near future. Despite good quantitative results, the main purpose of the present study remains the novel methodology that was set up for uncertainty propagation in CFD.