Faces are characterized by the simultaneous presence of several evaluation-relevant features, for example, emotional expression and (prejudiced) ethnicity. The social message account (SMA) hypothesizes the immediate integration of emotion and ethnicity. According to SMA, happy in-group faces should be interpreted as benevolent, whereas happy out-group faces should be interpreted as potentially malevolent. By contrast, fearful in-group faces should be interpreted as signaling an unsafe environment, whereas fearful out-group faces should be interpreted as signaling inferiority. In contrast, the processing conflict account (...) (PCA) assumes that each face conveys two rather independent evaluative features, emotion and ethnicity. Thus, stimuli might be either affectively congruent or incongruent, and thereby exert influences on behavior. The article reviews the evidence with regard to the two accounts before reporting an experiment that aims at disentangling them. In an approach/avoidance task (AAT), either happy/fearful faces of German and Turks were presented or happy/fearful faces of young and old persons. There are prejudices against Turk/Middle-eastern persons (in Germany) as well as against old persons. For SMA, the two prejudices are of different type; thus prediction for the AAT diverge for the two group conditions. In contrast, for PCA both group features (i.e., Turk ethnicity and old age) are negative features (in comparison to their counterparts) which are affectively congruent or incongruent to the emotional expression. Hence, the results pattern in the AAT should be comparable for the two group conditions. In accordance with SMA but in contrast to PCA, we found different patterns for the two group conditions. (shrink)