The term "cortical column" has been used to identify a number of different types of subdivisions of cortical areas. Here we describe and discuss several of these types of modular subdivisions in cortex, and suggest criteria for distinguishing these types. Minicolumns are narrow, vertical arrays of densely interconnected neurons that cross all cortical layers, and may represent a basic computational unit that is common to all cortical areas across mammalian species. Other types of columns are more variable across cortical areas and mammalian species, and have different developmental and evolutionary functions and histories. Classical columns are so named because they correspond to the column type first described by Mountcastle (J Neurophysiol 20:408-434, PMID 13439410, 1957), and consist of larger alternating patches of cortex that span all cortical layers, and contain neurons that have different response properties from those of neurons in adjoining columns. Several types of classical columns have been identified, and evidence for more types will likely emerge with further investigation. Classical columns divide cortical areas into sets of functionally specialized modules. A third type of cortical subdivision, unbounded columns, collectively represent a continuously varied stimulus dimension, such as the orientation of a bar or line, across a cortical area. Thus, vertical rows of neurons across a patch of cortex representing a small portion of the visual field will vary continuously in the orientation of a stimulus bar that best activates them, and there are no obvious borders between rows of neurons that prefer one orientation and adjoining rows that prefer a slightly different orientation. Other continuously variable stimuli such as color are likely to be represented by unbounded columns as well. A fourth type of cortical column consists of adjoining blocks of neurons that represent different parts of the receptor surface. The ocular dominance columns in primary visual cortex constitute a wellknown example, where blocks of neurons are activated preferentially by one eye or the other. The banded representation of digits in somatosensory cortex of monkeys and raccoons provide another example. Lastly, cortical areas are sometimes divided into specialized regions or domains that are distinct from one another and do not repeat across the cortical surface. Such domains appear to exist in motor and