After stroke, penumbral salvage determines clinical recovery. However, the rescued penumbra may be affected by selective neuronal loss, as documented both histopathologically in animals and using the validated in vivo positron emission tomography marker (11)C-flumazenil in humans. However, whether the non-infarcted penumbra is capable of neuronal activation, and how selective neuronal loss may interfere, is unknown. Here we prospectively mapped the topographical relationships between functional magnetic resonance imaging responses and non-infarcted penumbra, and tested the hypothesis that the former do take place in the latter, but only in its subsets spared selective neuronal loss. Seven patients (mean age 74 years; three thrombolysed) with first-ever acute anterior circulation stroke, presence of penumbra on computed tomography perfusion performed within 6 h of onset, and substantial deficit on admission but good outcome at 1-3 months (National Institute of Health Stroke Score range 6-13 and 0-1, respectively, P = 0.001), were studied. At follow-up, patients underwent structural magnetic resonance imaging to map the infarct, functional magnetic resonance imaging (three tasks selected to probe the right or left hemisphere), and (11)C-flumazenil positron emission tomography generating binding potential maps. Patients with significant carotid or middle-cerebral artery disease or impaired vasoreactivity were excluded. Following image coregistration, the non-infarcted penumbra comprised all acutely ischaemic voxels (identified on acute computed tomography perfusion using previously validated thresholds) not part of the final infarct. To test our hypotheses, the overlap between functional magnetic resonance imaging activation clusters and non-infarcted penumbra was mapped, and binding potential values then computed both within and outside this overlap. In addition, the overlap between functional magnetic resonance imaging activation clusters and areas of significantly reduced binding potential (determined using Statistical Parametric Mapping against 16 age-matched control subjects) was assessed in each patient. An overlap between non-infarcted penumbra and functional magnetic resonance imaging clusters was present in seven of seven patients, substantial in four. Binding potential was significantly reduced in the whole non-infarcted penumbra (P < 0.01) but not within the functional magnetic resonance imaging overlap. Clusters with significantly reduced binding potential showed virtually no overlap with functional magnetic resonance imaging activation compared with 12 age-matched controls (P = 0.04).The results from this proof of principle study suggest that 1-3 months after stroke the non-infarcted penumbra is capable of neuronal activation, consistent with its established role in recovery of neurological functions. However, although the non-infarcted penumbra as a whole was affected by selective neuronal loss, activations tended to occur within portions spared selective neuronal loss, suggesting the latter impedes neuronal activation. Although its clinical correlates are still elusive, selective neuronal loss may represent a novel therapeutic target in the aftermath of ischaemic stroke.