Global sensitivity analysis (GSA) is a well-established approach to support simulation-driven design decisions, where the dependency between the simulation’s output and the model’s input is quantified. However, classical GSA approach, such as Sobol’ indices based on Monte Carlo Simulations (MCS), is not convenient when computationally expensive simulation models such as Representative Volume Elements (RVE) are used. The study develops a simulation framework with a metamodeling-based GSA to overcome the aforementioned cost of the MCS approach. The developed framework has been applied in a multi-scale modelling framework replacing the RVE simulations, for CFRP material with three different metamodels for performing GSA. The objective is to predict how the manufacturing process and introduction of defects may influence the fatigue behavior of the material. A robust model to predict the fatigue life of these materials is in need and therefore it is important to determine how to consider defects in these simulation frameworks, and how defects may affect the fatigue life.