In artificial neural networks, backpropagation learning, do we have to worry about overftting?

Do we need to have a validation data set in addition do the training data set? do we have to perform something like cross-validation to avoid overfitting?