Air hunger (uncomfortable urge to breathe) is a component of dyspnea (shortness of breath). Three human H(2)(15)O positron emission tomography (PET) studies have identified activation of phylogenetically ancient structures in limbic and paralimbic regions during dyspnea. Other studies have shown activation of these structures during other sensations that alert the organism to urgent homeostatic imbalance: pain, thirst, and hunger for food. We employed blood oxygen level dependent (BOLD) functional magnetic resonance imaging (fMRI) to examine activation during air hunger. fMRI conferred several advantages over PET: enhanced signal-to-noise, greater spatial resolution, and lack of ionizing radiation, enabling a greater number of trials in each subject. Six healthy men and women were mechanically ventilated at 12-14 breaths/min. The primary experiment was conducted at mean end-tidal PCO(2) of 41 Torr. Moderate to severe air hunger was evoked during 42-s epochs of lower tidal volume (mean = 0.75 L). Subjects described the sensation as "like breath-hold," "urge to breathe," and "starved for air." In the baseline condition, air hunger was consistently relieved by epochs of higher tidal volume (mean = 1.47 L). A control experiment in the same subjects under a background of mild hypocapnia (mean end-tidal PCO(2) = 33 Torr) employed similar tidal volumes but did not evoke air hunger, controlling for stimulus variables not related to dyspnea. During each experiment, we maintained constant end-tidal PCO(2) and PO(2) to avoid systematic changes in global cerebral blood flow. Whole-brain images were acquired every 5 s (T2*, 56 slices, voxel resolution 3 x 3 x 3 mm). Activations associated with air hunger were determined using voxel-based interaction analysis of covariance that compared data between primary and control experiments (SPM99). We detected activations not seen in the earlier PET study using a similar air hunger stimulus (Banzett et al. 2000). Limbic and paralimbic loci activated in the present study were within anterior insula (seen in all 3 published studies of dyspnea), anterior cingulate, operculum, cerebellum, amygdala, thalamus, and basal ganglia. Elements of frontoparietal attentional networks were also identified. The consistency of anterior insular activation across subjects in this study and across published studies suggests that the insula is essential to dyspnea perception, although present data suggest that the insula acts in concert with a larger neural network.