Although the distinction between positive and negative facial expressions is assumed to be clear and robust, recent research with intense real-life faces has shown that viewers are unable to reliably differentiate the valence of such expressions (Aviezer, Trope, & Todorov, 2012). Yet, the fact that viewers fail to distinguish these expressions does not in itself testify that the faces are physically identical. In Experiment 1, the muscular activity of victorious and defeated faces was analyzed. Higher numbers of individually coded facial actions—particularly smiling and mouth opening—were more common among winners than losers, indicating an objective difference in facial activity. In Experiment 2, we asked whether supplying participants with valid or invalid information about objective facial activity and valence would alter their ratings. Notwithstanding these manipulations, valence ratings were virtually identical in all groups, and participants failed to differentiate between positive and negative faces. While objective differences between intense positive and negative faces are detectable, human viewers do not utilize these differences in determining valence. These results suggest a surprising dissociation between information present in expressions and information used by perceivers.