One of the practical problems in neuroimaging using near infrared spectroscopy (NIRS) is to choose an appropriate source-detector distance to maximize the sensitivity to cerebral blood oxygenation and to improve the spatial resolution for mapping cortical activation. While NIRS has attracted increasing attention in neuroimaging in infants, there has been no report of comparative data regarding source-detector distance for the infant brain. In the present study, 9 quietly sleeping 3-month-old infants were exposed to 3-s speech sounds, and hemodynamic responses over the bilateral temporal cortices were assessed by using multiple pairs of source and detector of NIR light with varying distances (1, 2, 3 and 4 cm) and varying intensities (0.6 and 1.2 mW). The statistical analyses of the group-averaged hemodynamic responses and the frequency analyses of the signal-to-noise ratios revealed that a 2-cm source-detector distance with 0.6-mW NIR light provided the highest sensitivity to cortical responses. This indicates that NIRS can be used to detect the activation of the cortical regions, in the infant brain, by using the source-detector distance scaled to the smaller head size of infants and a relatively low intensity of NIR light compared to the ones that have been routinely used in adult studies.