The human brain is highly complex, with billions of neurons and connections interacting and contributing to its final operation. Understanding the role of each single component is far from trivial. Aiming to exemplify the analysis of brain parts, different models have been proposed across years, with a variable degree of biological plausibility. Artificial neural networks represent a widespread and convenient tool, due to their biological inspiration. However, in most of the practical applications a certain number of assumptions and simplifications is made, and understanding the role of each element (e.g., neurons, connections) is not straightforward. Bearing this in mind, we investigate the importance of single components in the operation of an artificial neural network. We introduce the concept of functional network and explain its close relationship with the final performance. More specifically, we present a detailed analysis of different neural network topologies, where the number of internal layers and/or the number of neurons per layer is varied. We show how effective solutions typically rely on larger functional networks. We also demonstrate that complex controllers are less evolvable because their functional network is small. Our preliminary results have been obtained on a variant of the well-known foraging scenario involving an E-puck mobile robot.