In deep learning, variance refers to the variability or inconsistency of the model's performance when trained on different subsets of the training data. A high variance model is one that overfits to the training data and does not generalize well to unseen data.