Destruction of the occipital cortex presumably leads to permanent blindness in the contralateral visual field. Residual abilities to respond to visual stimuli in the blind field without consciously experiencing them have, however, been described in cortically blind patients and are termed ‘blindsight’. Although the neuronal basis of blindsight remains unknown, possible neuronal correlates have been proposed based on the nature of the residual vision observed. The most prominent but still controversial hypothesis postulates the involvement of the superior colliculi in blindsight. Here we demonstrate, using a computer-based reaction time test in a group of hemispherectomized subjects, that human ‘attention-blindsight’ can be measured for achromatic stimuli but disappears for stimuli that solely activate S-cones. Given that primate data have shown that the superior colliculi lacks input from S-cones, our results lend strong support to the hypothesis that ‘attention-blindsight’ is mediated through a collicular pathway. The contribution of a direct geniculo-extrastriate-koniocellular projection was ruled out by testing hemispherectomized subjects in whom a whole hemisphere has been removed or disconnected for the treatment of epilepsy. A direct retino-pulvinar-cortical connection is also unlikely as the pulvinar nucleus is known to receive input from S-cones as well as from L/M-cone-driven colour-opponent ganglion cells.