Background: Face processing in Williams syndrome (WS) has been a topic of heated debate over the past decade. Initial claims about a normally developing (‘intact’) face-processing module were challenged by data suggesting that individuals with WS used a different balance of cognitive processes from controls, even when their behavioural scores fell within the normal range. Measurement of evoked brain potentials also point to atypical processes. However, two recent studies have claimed that people with WS process faces exactly like normal controls. Method: In this paper, we examine the details of this continuing debate on the basis of three new face-processing experiments. In particular, for two of our experiments we built task-specific full developmental trajectories from childhood to adolescence/ adulthood and plotted the WS data on these trajectories. Results: The first experiment used photos of real faces. While it revealed broadly equivalent accuracy across groups, the WS participants were worse at configural processing when faces were upright and less sensitive than controls to face inversion. In Experiment 2, measuring face processing in a storybook context, the face inversion effect emerged clearly in controls but only weakly in the WS developmental trajectory. Unlike the controls, the Benton Face Recognition Test and the Pattern Construction results were not correlated in WS, highlighting the different developmental patterns in the two groups. Again in contrast to the controls, Experiment 3 with schematic faces and non-face stimuli revealed a configural-processing deficit in WS both with respect to their chronological age (CA) and to their level of performance on the Benton. Conclusion: These findings point to both delay and deviance in WS face processing and illustrate how vital it is to build developmental trajectories for each specific task.