Building animated conversational agents requires developing a fine-grained analysis of the motions and meanings available to interlocutors in face-to-face conversation and implementing strategies for using these motions and meanings to communicate effectively. In this paper, we sketch our efforts to characterize people's facial displays of uncertainty in face-to-face conversation. We analyze empirical data from human–human conversation and extend our platform for conversational animation, including RUTH (the Rutgers University Talking Head), to simulate what we find. This methodology leads to a range of new insights into the structure, timing, expressive content and communicative function of facial actions.