Visual scenes explored covertly are initially represented in a retinal frame of reference (FOR). On the other hand, ‘later’ stages of the cortical network allocating spatial attention most probably use non-retinal or non-eye-centred representations as they may ease the integration of different sensory modalities for the formation of supramodal representations of space. We tested if the cortical areas involved in shifting covert attention are based on eye-centred or non-eye-centred coding by using functional magnetic resonance imaging. Subjects were scanned while detecting a target item (a regularly oriented ‘L’) amidst a set of distractors (rotated ‘L's). The array was centred either 5° right or left of the fixation point, independent of eye-gaze orientation, the latter varied in three steps: straight relative to the head, 10° left or 10° right. A quantitative comparison of the blood-oxygen-level-dependent (BOLD) responses for the three eye-gaze orientations revealed stronger BOLD responses in the right intraparietal sulcus (IPS) and the right frontal eye field (FEF) for search in the contralateral (i.e. left) eye-centred space, independent of whether the array was located in the right or left head-centred hemispace. The left IPS showed the reverse pattern, i.e. an activation by search in the right eye-centred hemispace. In other words, the IPS and the right FEF, members of the cortical network underlying covert search, operate in an eye-centred FOR.