A prominent feature of the primate brain is the existence of several cortical areas in which neurons respond categorically to certain visual stimuli.  Most notably are the so-called "face patches", defined as regions in which fMRI responses are stronger to face images than to other types of visual objects. Recent advances in our laboratory have made it possible to record longitudinally from individual neurons across weeks and months, opening the door for new types of experiments to investigate face-selective neurons.  In my talk I will describe the results of several such experiments.  For example, we found that the visual selectivity of neurons in face patch AF, a "high-level" face patch, remain absolutely stable over a period up to a year, even during periods of intensive training on face identity.  In agreement with several studies, I will show that neurons in this area are highly selective for faces -- in this case, based on the presentation of 10,000 flashed stimuli over a period of 3 weeks. However, the same population of "face cells", when measured during free viewing of a natural video, bear a less obvious relationship to faces and diverge significantly in their responses.  Under these conditions, single neurons show a highly deterministic pattern of spiking upon repeated presentations of the same video stimulus.  However, neighboring neurons are often uncorrelated in their response time courses, reflecting a different functional specialization. I will show using a novel method, combining single-unit recordings and fMRI, that individual neurons fall into several distinct categories, each bearing a unique correlative relationship with functional networks throughout the brain.