Gaining a better understanding of the structure-property relationship in materials is a vital step in optimizing forming processes in order to minimize the forming-induced damage and thereby maximize the materials’ performance. Dual phase (DP) steels are composed of hard martensite surrounded by a soft and ductile ferrite matrix. Due to the complex microstructure of DP-steels, different mechanisms of damage initiation can occur, such as martensite cracking or ferrite-martensite phase boundary decohesion [1]. In a previous study, a multi-objective Bayesian optimization strategy was proposed for the design of damage-tolerant DP-microstructures, which optimizes the microstructure with respect to the two prominent damage initiation mechanisms in the utilized DP steel. It combined full-field crystal plasticity simulations on 3D representative volume elements (RVE) with computational optimization. One key problem with this approach is the substantial scatter in the simulation outcomes, especially when the calculations are done using RVE with a smaller edge length. However, simulations using RVE with a larger edge length are computationally more expensive than those with shorter edges. Since the optimization is done in several epochs, using the computationally more expensive approach results in significantly longer calculations. To tackle these issues, this work explores the possibility of using multi-fidelity Bayesian optimization, where the algorithm chooses between two models for each calculation. The first model uses RVE with shorter edges and is therefore less computationally expensive, while the second model employs RVE with longer edges that promise less scatter in the results. This leads to an improved performance in the optimization without excessive calculation times.