The amount by which a model's predictions change when trained on different datasets. High variance indicates the model is sensitive to small changes in training data, often leading to overfitting.