Experimental paradigm (from Gao et al. 2018). (A) During the scanning run, variable natural images of nonface stimuli are presented at a fast rate of 6 Hz (6 images/s). Every 9 s (0.111 Hz), a “mini-burst” of 7 highly variable face images alternates with nonface images, covering a period of 2.167 s. This procedure prevents category-based adaptation effects within a mini-block and, with a stimulus-onset asynchrony of 333 ms between two faces, allows capturing the bulk of every underlying face-selective response (Retter and Rossion 2016). These two factors contribute to the high signal-to-noise ratio of the face-selective activation observed with this approach (Gao et al. 2018). The frequency of face bursts (0.111 Hz) is referred to as the face stimulation frequency with a signal at 0.111 Hz in the fMRI spectrum, reflecting face-selective activity. (B) In one example brain, the highest face-selective activity at 0.111 Hz as found in the right lateral middle fusiform gyrus, with no increased 0.111 Hz activity to the same images matched for low-level properties (amplitude spectrum) but phase-scrambled to prevent face or object recognition.