Any theory of the neural mechanisms of language requires an understanding of the cerebral cortex, for this organ undoubtedly plays a part in the production and perception of speech. The connectivity of the cortical network, in our opinion, is best described in statistical terms, since no suggestion of specific "wiring" of individual neurons in the cortex has ever gone beyond speculation. The statistical picture is that of a huge network of elements of one kind, the pyramidal cells, connected to each other by a special kind of synapse, residing on dendritic spines. These synapses are very numerous, very weak, excitatory and probably plastic. These four properties make it quite likely that the cortex is a structure which stores memory in an associative way. Also, such a structure would be ideally suited for embodying information in the form of cell assemblies, as suggested by Hebb. Bearing in mind this picture of the cortex, which is now widely accepted, it is intruiguing to imagine what happens there during the production and comprehension of language. One clue is provided by the much-studied visual areas where (thanks to the work of Hubel and Wiesel) it is now quite obvious that individual neurons do not correspond to anything on the level of morphemes in language, but rather signal events at the level of phonemes or probably even below. The generation of syllables is very likely a process involving large sets of neurons, and the chaining of syllables into sentences probably requires the cooperation of several other parts of the brain in addition to the cortex.