The paper proposes a population-based meta-heuristic approach for the multi-objective robust optimization of tooth profiles, aimed at finding a set of micro-geometry modification parameters that allow to improve mechanical performance of spur gears. With the aim of making the optimization results reliable in real-life applications, a robust formulation of the optimization problem is generated by incorporating noise parameters that account for the influence of manufacturing uncertainties on the objective function. The described multi-objective gear optimization strategy is based on response surface (surrogate) models, allowing for checking the performance of a large number of candidate solutions in very short computational times. The computation efficiency of the proposed approach is the key that enables a simultaneous assessment of both linear and parabolic profile modifications, so that the most appropriate tooth geometry can be selected for the specific application. The proposed approach was successfully employed in a case study in which static transmission error and contact stress of a gear pair loaded by different torques were optimized under fatigue-related constraints and in presence of geometric variability due to manufacturing uncertainties.