Faces with typically African features are perceived as darker than they really are. We investigated how early in processing the bias emerges, whether participants are aware of it, and whether it can be altered by explicit instructions. We presented pairs of faces sequentially, manipulated the luminance and morphological features of each, and asked participants which was lighter, and how confident they were in their responses. In Experiment 1, pre-response mouse cursor trajectories showed that morphology affected motor output just as early as luminance did. Furthermore, participants were not slower to respond or less confident when morphological cues drove them to give a response that conflicted with the actual luminance of the faces. However, Experiment 2 showed that participants could be instructed to reduce their reliance on morphology, even at early stages of processing. All stimuli used, code to run the experiments reported, raw data, and analyses scripts and their outputs can be found at https://osf.io/brssn.