Feature Selective Neuroevolution of Augmenting Topologies (FS-NEAT) and Feature Deselective Neuroevolution of Augmenting Topologies (FD-NEAT) are two popular methods for optimizing the topology and the weights of Artificial Neural Networks (ANNs) simultaneously with performing feature selection. However, no study exists that systematically investigates their performance on the exclusive-or (XOR) problem with increasing complexity. Moreover, it is unknown whether the choice of a different initial topology of the ANNs would influence the performance of the two algorithms. For this reasons, this paper investigates the performance of FD-NEAT and FS-NEAT in terms of accuracy, number of generations required for their convergence to the optimal solution and their ability of selecting the relevant features in artificial datasets with irrelevant features. The comparisons are performed based on hypothesis tests (Wilcoxon rank sum test, p<0.05). The results show that the choice of the initial topology can affect the performance of the two algorithms, resulting in higher accuracy, faster convergence and better feature selection abilities.