Neuroevolution techniques (NE) were developed to address the challenge of subjective definition in the architecture and topology of artificial neural networks. They draw inspiration from the biological process of evolution, iteratively searching for the best solutions among generations of networks. In this work, we performed an exhaustive analysis to evaluate the relevance of the numerous parameters involved in NE models and assess their impact on the performance of the techniques. This study assesses the performance of NEAT and HyperNEAT techniques to determine their capability to develop the required structures for solving a given problem. Specifically, the primary objective is to ascertain whether the topology construction takes place reliably. To ensure comprehensive coverage of the sample space, we employed sampling and statistical inference methods. We obtained independent observations for performing statistical significance tests and analyzing the performance of the NE techniques under study. Results of several experiments show how NEAT tends to employ a smaller number of neurons and links, while HyperNEAT exhibits a notably higher architecture complexity.