The relationship between location and extent of lesion on CT scan and limitation in spontaneous speech was examined. The severity of spontaneous speech ranged from cases with no speech or only verbal stereotypies (first major group) to those with reduced, hesitant, poorly articulated, agrammatic speech (nonfluent Broca's aphasia, second major group). CT scan analysis revealed no single neuroanatomical area that contained an extensive lesion which could be used to discriminate the most severe cases from the least severe. The two groups were separable, however, on the basis of the CT scan when the extent of the lesion in two subcortical white matter areas were combined: (1) the most medial and rostral portion of the subcallosal fasciculus plus (2) the periventricular white matter near the body of the lateral ventricle, deep to the lower motor/sensory cortex area for the mouth. The most rostral portion of the medial subcallosal fasciculus, located in the lateral angle of the frontal horn (extremely deep to Broca's area), contains projections from the cingulate gyrus (area 24) and the supplementary motor area, to the caudate nucleus. We suggest that one explanation for the more severe limitation in spontaneous speech in the first group is the extensive white matter lesion in these two subcortical pathways had interrupted a large number of connections for (1) initiation and preparation of speech movements, and limbic aspects of speech (lesions in the medial subcallosal fasciculus), and (2) motor execution and sensory feedback for spontaneous speech (lesions in periventricular white matter deep to the motor/sensory cortex area for the mouth). Extensive lesion in only one of these two white matter pathway areas, alone was not sufficient to produce long-lasting severe limitation in spontaneous speech and could not be used to discriminate the two groups on the basis of the CT scans. The patients with less severe limitation in spontaneous speech (nonfluent Broca's aphasia) had less extensive lesion within these two white matter areas combined, and had interrupted a smaller number of these subcortical connections. The sites of the lesions in subcortical white matter in CT scans in Broca's original case who could only produce a verbal stereotypy are similar to those in our first group with the most severe limitation in spontaneous speech. The presence or absence of hemiplegia was not related to severity or recovery of spontaneous speech. Careful examination of lesion extent in these two areas of subcortical white matter on CT scanning appears to be relevant in predicting potential for recovery of spontaneous speech in some stroke patients.