Overfitting is when a statistical model learns patterns in the training data that are too complex to generalize well.