We critically review the literature on subcortical aphasia, suggest that a number of traditional concepts regarding mechanisms of aphasia are inconsistent with now abundant data, and propose several new hypotheses. The absence of aphasia in 17 reported cases of dominant hemisphere striatocapsular infarction and the finding of nearly every conceivable pattern of language impairment in 33 different reported cases of striatocapsular infarction provide strong evidence against a major direct role of the basal ganglia in language and against disconnection or diaschisis as mechanisms of nonthalamic subcortical aphasia. However, detailed consideration of the vascular events leading to striatocapsular infarction strongly suggests that associated linguistic deficits are predominantly related to sustained cortical hypoperfusion and infarction not visible on structural imaging studies. Thalamic disconnection, as may occur with striatocapsular infarcts with extension to the temporal stem and putamenal hemorrhages, may also contribute to the language deficits in some patients. Review of the literature on thalamic infarction, in conjunction with previously unreported anatomic details of four cases, suggests that what infarcts in the tuberothalamic artery territory and the occasional infarcts in the paramedian artery territory associated with aphasia have in common is damage to the frontal lobe-inferior thalamic peduncle-nucleus reticularis-center median system that may be involved in regulating the thalamic gate in attentional processes. Disruption of attentional gating in the pulvinar and lateral posterior nuclei resulting from such lesions may impair selection of specific neuronal networks in the projection field of these nuclei that serve as the substrate for lexical-semantic function, which is in effect a disruption of a type of working memory, as defined by Goldman-Rakic. We define this as a defect of selective engagement.