While much is known about the appearance and human perception of emotional facial expressions, researchers and professionals experience difficulties when attempting to create believable animated characters. Methods for automating or capturing dynamic facial expressions have come on in leaps and bounds in recent years, resulting in increasingly realistic characters. However, accurate replication of naturalistic movement does not necessarily ensure authentic character performance. In this paper, the authors present a project which makes use of creative animation practices and artistic reflection as methods of research. The output of animation practice is tested experimentally by measuring observer perception and comparing the results with artistic observations and predictions. Ultimately, the authors aim to demonstrate that animation practice can generate new knowledge about dynamic character performance, and that arts-based methods can and should be considered valuable tools in a field often dominated by technical methods of research.