Experiments were performed to examine the topography of covert visual attention signals in human superior colliculus (SC), both across its surface and in its depth. We measured the retinotopic organization of SC to direct visual stimulation using a 90° wedge of moving dots that slowly rotated around fixation. Subjects (n = 5) were cued to perform a difficult speed-discrimination task in the rotating region. To measure the retinotopy of covert attention, we used a full-field array of similarly moving dots. Subjects were cued to perform the same speed-discrimination task within a 90° wedge-shaped region, and only the cue rotated around fixation. High-resolution functional magnetic resonance imaging (fMRI, 1.2 mm voxels) data were acquired throughout SC. These data were then aligned to a high-resolution T1-weighted reference volume. The SC was segmented in this volume so that the surface of the SC could be computationally modeled and to permit calculation of a depth map for laminar analysis. Retinotopic maps were obtained for both direct visual stimulation and covert attention. These maps showed a similar spatial distribution to visual stimulation maps observed in rhesus macaque and were in registration with each other. Within the depth of SC, both visual attention and stimulation produced activity primarily in the superficial and intermediate layers, but stimulation activity extended significantly more deeply than attention.