Abstract Whether visual categorization, i.e., specific responses to a certain class of visual events across a wide range of exemplars, is graded or all-or-none in the human brain is largely unknown. We address this issue with an original frequency-sweep paradigm probing the evolution of responses between the minimum and optimal presentation times required to elicit both neural and behavioral face categorization responses. In a first experiment, widely variable natural images of nonface objects are progressively swept from 120 to 3 Hz (8.33 to 333 ms duration) in rapid serial visual presentation sequences; variable face exemplars appear every 1 s, enabling an implicit frequency-tagged face-categorization electroencephalographic (EEG) response at 1 Hz. In a second experiment, faces appear non-periodically throughout such sequences at fixed presentation rates, while participants explicitly categorize faces. Face-categorization activity emerges with stimulus durations as brief as 17 ms for both neural and behavioral measures (17 – 83 ms across individual participants neurally; 33 ms at the group level). The face-categorization response amplitude increases until 83 ms stimulus duration (12 Hz), implying graded categorization responses. However, a strong correlation with behavioral accuracy suggests instead that dilution from missed categorizations, rather than a decreased response to each face stimulus, may be responsible. This is supported in the second experiment by the absence of neural responses to behaviorally uncategorized faces, and equivalent amplitudes of isolated neural responses to only behaviorally categorized faces across presentation rates, consistent with the otherwise stable spatio-temporal signatures of face-categorization responses in both experiments. Overall, these observations provide original evidence that visual categorization of faces, while being widely variable across human observers, occurs in an all-or-none fashion in the human brain.