Using a yes/no task in which distracters were closely matched to targets, three experiments examined the effect of encoding operations on recognition memory for upright and inverted faces. Correct recognition for targets (hits) was lower on inverted than on upright faces. However, the decline was smaller for feature than for trait judgments or for expression judgments (Experiment 1), for feature than for trait judgments (Experiment 2), and for feature than for global judgments (Experiment 3). These findings suggest that faces are primarily encoded in terms of isolated components with feature judgments, and in terms of configural information (spatial information among features) with trait, expression, and global judgments.