Social group identity plays a central role in political polarization and inter-party conflict. Here, we use ambiguously valenced faces to measure bias in the processing of political ingroup and outgroup faces, while also accounting for inter-party differences in judgments of emotion at baseline. Participants identifying as Democrats and Republicans judged happy, angry, and surprised faces as positive or negative. Whereas happy and angry faces convey positive and negative valence respectively, surprised faces are ambiguous in that they readily convey positive and negative valence. Thus, surprise is a useful tool for characterizing valence bias (i.e., the tendency to judge ambiguous stimuli as negative). Face stimuli were assigned to the participants' political ingroup or outgroup, or a third group with an unspecified affiliation (baseline). We found a significant interaction of facial expression and group membership, such that outgroup faces were judged more negatively than ingroup and baseline, but only for surprise. There was also an interaction of facial expression and political affiliation, with Republicans judging surprise more negatively than Democrats across all group conditions. However, we did not find evidence for party differences in outgroup negativity. Our findings demonstrate the utility of judgments of surprised faces as a measure of intergroup bias, and reinforce the importance of outgroup negativity (relative to ingroup positivity) for explaining inter-party biases.