In this paper, we explore the roles of neurons and layers in artificial neural networks (ANNs) designed to classify one-dimensional inputs into three distinct classes over the real line R. The classes are defined as (−∞, − 1], (−1, 1], and (1, ∞). We train a neural network with a single input neuron, a hidden layer of three ReLU-activated neurons, and an output layer of three softmax-activated neurons corresponding to the classes. Using visualisation techniques within the NNVisualiser framework, we analyse how the network transforms inputs and makes classification decisions. Our findings reveal that while the hidden layer introduces nonlinearity essential for complex decision boundaries, a simpler model without a hidden layer suffices for linearly separable data. This experimentation aids in explaining the inner workings of ANNs as classifiers, contributing to the broader field of explainable artificial intelligence (XAI).