Manipulation of generality, through appropriate generalization and specialization, can modify classifier performance in predictable and useful ways.