In this commentary, I will focus on two critical questions that must be addressed by any constructivist or experience-expectant account of the development of face processing. These can be thought of as the ‘what’ and ‘where’ questions. The ‘what’ question concerns why infants’ brains specialize more for faces than for other dynamic multi-modal objects in their environment. The ‘where’ question addresses why it is that some specific regions of cortex, such as the ‘fusiform face area’ (FFA), become specialized for faces and not, usually, other regions. Together, these issues result in what I have termed the ‘paradox of plasticity’ (Johnson et al., 1998). While I agree completely with Nelson’s conclusion that, overall, the available data better fit a constructivist than a nativist account of the neurodevelopment of face processing, we have to acknowledge that, from a nativist perspective, the what and where questions are answered in an attractively simple way: the brain is specialized for faces because particular genes combine to ‘code’ for specific neural wiring in a specific part of the brain. This specific part of the brain is assumed to be active early, and matures over the first months with, perhaps, a little fine tuning from experience. Nelson’s alternative answer to the ‘what’ question is simply that, in their natural early environment, babies are exposed to faces more frequently than to other objects, and so this class of visual stimuli shapes neural wiring from the earliest minutes. Specifically, Nelson criticizes evidence for a ‘Conspec’ (Johnson and Morton, 1991) on the grounds that (a) ‘faces are far from the only stimulus that moves . . . in the periphery, and thus it is not clear why Conspec would be positively biased toward faces’, (b) there is no evidence that (subcortical) visual motor pathways respond to patterned stimuli such as faces, and (c) while face preferences have been tested as early as the first half an hour of life (e.g. Johnson et al., 1991), they have never been tested at the actual point of birth, leaving open the possibility of very rapid experience-driven effects. Point (a) seems to be an unfortunate misunderstanding of the notion of a Conspec. Information about the approximate spatial arrangement of high-contrast elements that compose a face is necessary for Conspec precisely because there are many non-face objects in the infant’s early environment. Thus, this is an argument for, not against, a Conspec mechanism. With regard to point (b), it has recently been suggested that the pulvinar may be an important substrate for the