Discriminative power refers to the ability of a statistical or machine learning model to distinguish between different classes or outcomes effectively. It is crucial for evaluating the performance of classification models, as higher discriminative power indicates better accuracy and reliability in predictions.