The time of appearance of the search array was used as the onset time of the different events. Regressors representing estimated head movements (translation and rotation with six degrees of freedom) were added into the model as covariates of no interest to account for artefacts due to possible head

movements during scanning. In a first step, the linear contrast for all four search events vs all control conditions, i.e. [sR(fC), sL(fC), sL(fR), sR(fL)] > [‘all control conditions;] was calculated [sR(fC) = search right/fixate centre; sL(fC) = search left/fixate centre; sL(fR) = search left/fixate right; and sR(fL) = search right/fixate left; see Fig.  1A] for see more each subject to delineate the cortical areas involved in the covert search without any spatial bias. Significant changes were assessed using t-statistics. Single-subject contrast images (P < 0.001, 40 voxels) were analysed at the group level using a random-effect model. This analysis compares the average activation for a given CDK activation voxel with the variability of that activation over the estimated

population (Friston et al., 1999). Group-averaged activation is only reported if P < 0.001 and if more than 40 contiguous voxels are included in the cluster. In addition, for each subject we compared the individual search condition with its corresponding control condition, i.e. [sR(fC) > cR(fC)], [sL(fC) > cL(fC)], [sL(fR) > cL(fR)] and [sR(fL) > cR(fL)] in order to assess

the whole-brain pattern of BOLD activity accompanying covert shifts of attention for the condition of interest. Specifically, a random-effect analysis for a certain contrast was performed to determine activations, which were consistent across all subjects. The resulting statistical maps were corrected for multiple comparisons using a cluster size/z-threshold algorithm (Forman et al., 1995). Group-averaged activation is only reported if P < 0.001 and if unless more than 40 contiguous voxels are included in the cluster. The reported clusters passed correction for multiple comparisons by applying a false discovery rate (FDR) criterion of 0.005 at the voxel level (Table 1), except for the left FEF. These activations were projected on an SPM-averaged co-registered T1-weighted image. Using the toolbox MarsBar (Brett et al., 2002), the clusters obtained from the group-averaged significant voxel-wise t-map were defined as region of interest (ROI). In the next step, we wanted to identify the existence of voxels in the aforementioned parieto-frontal areas that would encode covert search in eye-centred coordinates, thus showing higher responses for eye-centred contralateral conditions than ipsilateral.