Near-infrared spectroscopy (NIRS) has the potential for clinical application in neuropsychiatry because it enables non-invasive and convenient measurement of hemodynamic response to cognitive activation. Using 24-channel NIRS in 12 healthy men, we examined the replicability of oxy- and deoxy-hemoglobin concentration ([oxyHb], [deoxyHb]) changes in the prefrontal cortex during the category fluency task over four repeated sessions (each 1-week apart). Multiple methods were employed to evaluate the replicability of magnitude, location, and time course of the NIRS signals ([oxyHb], [deoxyHb]). Task performances did not differ significantly across sessions, nor were they significantly correlated with NIRS signals. Repeated measures ANOVA and variance component analysis indicated high replicability of magnitude for both NIRS measures, whereas the effect sizes of between-session differences in [oxyHb] were not negligible. The number and spatial location of significantly activated channels were sufficiently replicable for both measures, except that the across-session overlap of significantly activated channels was weak in [deoxyHb]. The time course of the activation was acceptably replicable in both measures. Taken together, these findings suggest there is considerable replicability of multiple-time measurements of prefrontal hemodynamics during cognitive activation in men. Further studies using different conditions or assessing sensitivity to longitudinal changes following interventions are necessary.