Distributed Hierarchical Processingin the Primate Cerebral CortexDaniel J. Felleman1 and David C. Van Essen21 Department of Neurobiology and Anatomy,University of Texas Medical School, Houston, Texas77030, and 2 Division of Biology, CaliforniaInstitute of Technology, Pasadena, California 91125In recent years, many new cortical areas have beenidentified in the macaque monkey. The number of iden-tified connections between areas has increased evenmore dramatically. We report here on (1) a summary ofthe layout of cortical areas associated with vision andwith other modalities, (2) a computerized database forstoring and representing large amounts of informationon connectivity patterns, and (3) the application of thesedata to the analysis of hierarchical organization of thecerebral cortex. Our analysis concentrates on the visualsystem, which includes 25 neocortical areas that arepredominantly or exclusively visual in function, plus anadditional 7 areas that we regard as visual-associationareas on the basis of their extensive visual inputs. Atotal of 305 connections among these 32 visual andvisual-association areas have been reported. This rep-resents 31 % of the possible number of pathways if eacharea were connected with all others. The actual degreeof connectivity is likely to be closer to 40%. The greatmajority of pathways involve reciprocal connections be-tween areas. There are also extensive connections withcortical areas outside the visual system proper, includingthe somatosensory cortex, as well as neocortical, tran-sitional, and archicortical regions in the temporal andfrontal lobes. In the somatosensory/motor system, thereare 62 identified pathways linking 13 cortical areas,suggesting an overall connectivity of about 40%. Basedon the laminar patterns of connections between areas,we propose a hierarchy of visual areas and of somato-sensory/motor areas that is more comprehensive thanthose suggested in other recent studies. The currentversion of the visual hierarchy includes 10 levels ofcortical processing. Altogether, it contains 14 levels ifone includes the retina and lateral geniculate nucleusat the bottom as well as the entorhinal cortex and hip-pocampus at the top. Within this hierarchy, there aremultiple, intertwined processing streams, which, at alow level, are related to the compartmental organizationof areas VI and V2 and, at a high level, are related tothe distinction between processing centers in the tem-poral and parietal lobes. However, there are some path-ways and relationships (about 10% of the total) whosedescriptions do not fit cleanly into this hierarchicalscheme for one reason or another. In most instances,though, it is unclear whether these represent genuineexceptions to a strict hierarchy rather than inaccuraciesor uncertainties in the reported assignment.During the past decade, there has been an explosionof information about the organization and connectiv-ity of sensory and motor areas in the mammalian ce-rebral cortex. Many laboratories have concentratedtheir efforts on the visual cortex of macaque monkeys,whose superb visual capacities in many ways rivalthose of humans. In this article, we survey recentprogress in charting the layout of different corticalareas in the macaque and in analyzing the hierarchicalrelationships among these areas, particularly in thevisual system.The original notion of hierarchical processing inthe visual cortex was put forward by Hubel and Wiesel(1962, 1965) to account for a progressive increase inthe complexity of physiological receptive field prop-erties in the cat visual cortex. In particular, they sug-gested that a serial, feedforward scheme could ac-count for the generation of simple cells from LGNinputs, and complex cells, in turn, from simple cells.Likewise, the properties of hypercomplex cells andeven "higher-order hypercomplex cells" were attrib-uted to inputs from their immediate predecessors.However, the pure form of this hypothesis is difficultto reconcile with the finding of highly reciprocal con-nectivity and parallel channels discovered in morerecent studies of the visual pathway (cf. Rockland andPandya, 1979; Stone et al., 1979; Lennie, 1980; Lennieet al., 1990; Shapley, 1990). On the other hand, thereis no a priori reason to restrict the notion of hierar-chical processing to a strictly serial sequence. In gen-eral, any scheme in which there are well-defined lev-els of processing can be considered hierarchical.The notion that anatomical criteria could be usedto delineate a hierarchy of cortical areas first receiveddetailed scrutiny about a decade ago (Rockland andPandya, 1979; Friedman, 1983; Maunsell and Van Es-sen, 1983). Since this hypothesis was last reviewedsystematically (Van Essen, 1985), the number of iden-tified visual areas and identified connections has in-creased greatly. In addition, 2 recent studies (Ander-sen etal., 1990;Boussaoudetal., 1990) have proposedhierarchical relationships for parietal, temporal, andfrontal areas that are largely, but not completely, con-sistent with one another and with our previousschemes. Here, we provide a critical examination ofthe degree to which the entire ensemble of availabledata fits into an overall hierarchical scheme. We alsoreview the evidence that the principle of hierarchicalCerebral Cortex Jan/Feb 1991;1:1—47; 1047-3211/91/»2.00organization applies to other functional modalitiesand to other species besides macaques.A related theme in our analysis concerns the natureof concurrent processing streams in the visual cortex.These streams are linked at the input side to specificsubcortical inputs from the magnocellular (M) andparvocellular (P) layers of the LGN (cf. Blasdel andLund, 1983; Hubel and Livingstone, 1987) and, at theoutput side, to functionally distinct regions of theparietal and temporal lobes (Ungerleider and Mish-kin, 1982; Desimone and Ungerleider, 1989). Ouranalysis will emphasize that, on the one hand, thereis considerable segregation of information flowthroughout the visual pathway; on the other hand,there is also substantial intermixing and cross talkbetween streams at successive stages of processing.It is likely that these complexities in the anatomicalcircuitry reflect the multiplicity of computationalstrategies needed for efficient visual function (DeYoeand Van Essen, 1988).Subdivisions and Interconnections of the Visual CortexA Cortical MapOur primary format for illustrating the location ofdifferent visual areas involves the use of 2-D