# neural network output layer for binary classification

I'm using a neural network for a binary classification problem. Is it better to have one neuron in the output layer or to use two, i.e. one for each class?