Hemispatial neglect, characterized as failure to attend to contralesional space, is hypothesized by current neuroanatomical models to result from damage to a network involving the frontal, parietal, and cingulate cortices, basal ganglia, and thalamus. This study investigated this model of neglect in 81 right hemisphere-damaged acute stroke patients using 99mTc-HMPAO single photon emission-computed tomography (SPECT). In order to exploit the inherent collinearity of SPECT regional brain ratios, a novel statistical technique, partial least squares (PLS), was utilized. It makes use of high correlations to identify biologically relevant patterns of brain activity. Averaged ipsilesional cerebellar ratios from 152 brain segments were covaried with performance on subtests of the Sunnybrook Neglect Battery. In this patient sample, the most influential region identified by PLS corresponded to the area surrounding the right temporal-parietal-occipital (TPO) junction that included the right lateral occipital, temporal, and inferior parietal lobes. Hypoperfusion in the medial frontal cortex, including the anterior cingulate, also emerged as significantly associated with more severe neglect. Thus, hypoperfusion in only two of the five hypothesized network regions emerged as significantly associated with hemispatial neglect on SPECT imaging. This work converges with structural imaging studies to suggest that damage to the TPO junction, not just the parietal lobe, may be the critical region for hemispatial neglect. Our study demonstrated the utility of PLS for analyzing functional imaging and behavioral data sets in a clinical population in relation to current neuroanatomical models of neglect.