Somewhere on the ventral surface of the temporal lobe, in a region called the fusiform gyrus, there is a small area of cortex that responds far more strongly to faces than to almost anything else. The area is known as the fusiform face area, and two papers published roughly six years apart have each added something precise and somewhat counterintuitive to what we know about it.
The first finding, from a 2014 paper in Nature Communications, concerns timing. Using electrodes placed directly on the fusiform face area of four patients undergoing epilepsy monitoring, neuroscientist Avniel Singh Ghuman and colleagues at the University of Pittsburgh recorded electrical activity while participants viewed images of faces, bodies, houses, hammers, shoes, and phase-scrambled faces. What Ghuman’s team found was that the region was responding selectively to faces within 50 to 75 milliseconds of a face appearing on screen. That is faster than this region had previously been shown to respond to any visual category in humans, and faster than most other categories reach the temporal cortex at all.
The second finding, from a 2020 paper in the Proceedings of the National Academy of Sciences by N. Apurva Ratan Murty, Nancy Kanwisher, and colleagues at MIT, concerns development. When people who have been blind since birth handle three-dimensional printed models of faces, the fusiform face area becomes active. Visual experience with faces, it turns out, is not what builds the area’s preference for them.
These are findings from two separate papers, each with its own design, sample, and limits. Neither should be read as a settled account of how face perception works. But together they describe something worth sitting with.
What the Ghuman paper actually measured
The electrode method Ghuman’s team used is called intracranial electrocorticography, or ECoG. It involves recording directly from the brain’s surface at very high temporal resolution, far finer than what fMRI allows. The four participants in the study were epilepsy patients who already had electrodes implanted as part of their clinical care. The researchers used a machine learning algorithm to decode, on a trial-by-trial basis, whether the brain signal from the fusiform face area at any given moment was consistent with the participant viewing a face.
Face-selective activity appeared in the 50-to-75-millisecond window after stimulus onset and remained distinguishable from responses to other categories through to about 350 milliseconds. Crucially, phase-scrambled faces, which preserve the spatial frequency structure of a face but destroy its recognisable shape, did not produce the same early signal. This argues against the early response being driven simply by the visual statistics of a face-shaped image rather than by something more specific to face recognition.
The study was conducted on four participants. That is a small sample by the standards of most research, though the intracranial recording method compensates partly for sample size with signal precision that non-invasive imaging cannot approach. The authors are careful about what they claim: the early activity shows that face-specific information is present in the fusiform face area at 50 to 75 milliseconds, and they argue this is consistent with the region playing a role in initial face detection. They do not claim to have resolved all debates about the temporal architecture of face perception. There is ongoing disagreement in the field about when and where in the brain face selectivity first arises, and this paper contributes to that debate rather than closing it.
Beyond face detection, the same paper also found that the fusiform face area encodes which specific face someone is viewing, but this individuation happened considerably later, between 200 and 500 milliseconds, and was stable across changes in facial expression. And a late-sustained signal, broadband gamma activity lasting more than 500 milliseconds, tracked how long it took participants to respond in a gender-classification task. Longer gamma activity corresponded to slower responses. The area appears to be doing several different things at different moments, and not all at once.
A feature the area may not need to acquire
The MIT study took a different approach to a different question. Kanwisher and her colleagues wanted to know whether the fusiform face area develops its preference for faces because people spend years looking at faces, or whether the region has something more like a predetermined role that does not depend on that visual history.
To test this, they recruited people who had been blind from birth and had therefore never had visual experience with faces or anything else. Using fMRI, they scanned participants while they handled 3D-printed objects including faces, hands, chairs, and mazes. The fusiform face area was active during face handling, in roughly the same location it occupies in sighted people, and the selectivity for faces over other objects was comparable.
The finding does not mean visual experience is irrelevant to how the area functions in sighted people. Kanwisher has been quoted in MIT News as saying precisely that: visual input probably does play a role in sighted subjects. What the study shows is that visual experience is not necessary for the area to develop face selectivity in the first place. The researchers propose that long-range connectivity, the area’s structural relationships to other parts of the brain, may be what positions it to become selective for faces regardless of the sensory route through which face information arrives.
This finding builds on earlier work. A 2017 study from researchers in Belgium, published in the Proceedings of the National Academy of Sciences, scanned congenitally blind participants while they listened to face-related sounds such as laughing or chewing, and found elevated activity in the vicinity of the fusiform face area compared to non-face sounds. The MIT paper extended this with the more direct test of haptic face recognition.
What these two findings put together
Reading these papers alongside each other draws out something the standard account of the fusiform face area tends to flatten. It is easy to describe the area as a face-recognition module and leave it there. But Ghuman’s data show it operating at least three distinct processing stages, on three different timescales, doing different things with face information at different moments. And Murty and Kanwisher’s data show the area claiming its face selectivity without any visual faces ever having been seen.
What the area appears to have is something like a structural commitment to faces as a category, one that exists prior to, and independent of, a lifetime of looking. That does not mean it is a rigid or fixed processor. The late gamma activity Ghuman’s team found appears tied to working memory and task demands, suggesting the area is also responsive to what someone is trying to do with a face, not only to the presence of one.
The question of what this means for people with face recognition difficulties, or for understanding how face perception varies across individuals, is not something either paper directly addresses. Neither is clinical in that sense. Both are asking about the fundamental architecture of a region, not about what goes wrong when it malfunctions.
The limits worth naming
Ghuman et al. worked with four participants. That is small. The electrode placement was determined by clinical need, not experimental design, which means the precise location varied across subjects. The authors acknowledge their method is more sensitive to information encoded in temporal patterns than to information encoded spatially, so absence of a signal in their analysis does not necessarily mean absence of processing in the region.
The MIT haptic study relied on fMRI, which captures activity integrated over seconds rather than milliseconds. It tells us that the fusiform face area is involved when blind people handle face-shaped objects; it does not tell us precisely what computations are occurring or at what speed. The researchers are also working with a small group of congenitally blind participants, a population that is difficult to recruit in large numbers. The finding is worth taking seriously, but replication and extension will matter.
The broader question about what specifies the location and function of cortical areas is active and not resolved. The connectivity hypothesis is plausible and consistent with the data; it is not yet a confirmed account of development. Future work may complicate or qualify it.
What the research does not do
Neither paper suggests that the 50-millisecond finding represents the full story of how quickly the brain begins to recognise a face. Other regions contribute to face perception, and the signal Ghuman’s team recorded is from one area of the processing network. The fusiform face area appears to be unusually fast compared to how quickly the temporal cortex responds to non-face categories, but the authors are careful to frame this as a finding from this dataset with this method, not a universal statement about the speed of human face recognition.
Similarly, the MIT result about blindness does not mean the fusiform face area functions identically in people who have and have not had visual experience. Kanwisher says explicitly that visual input probably does shape the area in sighted people; the study only shows that such input is not required for face selectivity to emerge. These are different claims, and the difference matters.
The region continues to be studied, and the picture that has accumulated is more layered than a simple face-recognition box in the temporal lobe. It is active very early, it handles individual faces later, it maintains information in support of decisions, and it manages to do all of this without requiring the owner ever to have seen a face. How those capacities fit together is still being worked out.
