The common definition for robust solutions considers a solution robust if it remains optimal (or near optimal) when the parameters defining the fitness function are perturbed. We call this parameter robustness. In this thesis we propose an alternate definition for robustness, which we call solution robustness, if both the solution and the neighbourhood around the solution has high fitness. With this new definition, we created a set of functions with useful properties to allow for the testing of solution robustness. The performance of a Genetic Algorithm (GA) is then evaluated based on its ability to identify multiple robust solutions based on the solution robustness definition. It was found that the type of crossover and chromosome encoding scheme has no effect on the robustness of solution, while the sampling size (precision) of the search space does have a drastic effect on both the number of solutions and their quality. Furthermore, it was found that the commonly used niching approaches improve the multimodal aspect of robustness, but at the expense of the quality of the solutions found. Different neighbourhood evaluation schemes are also compared and of them the minimum neighbour technique was found to be the most effective. Based on these findings, a new robust multi-objective GA is proposed. This approach leads to the finding of more robust solutions than when single objective robust techniques were used. Finally, an in-depth investigation of solution robustness provides evidence of a close relationship between it and parameter robustness.