Currently, there is no widely used method to assess the reliability of contact between optodes and tissue in near-infrared spectroscopy (NIRS). In this study we observe a high linear dependence (R(2) approximately 0.99) of the logarithmic modulation amplitude (ln(I(AC))), average intensity (ln(I(DC))) and phase (phi) on the source-detector distance (SDD) ranging from approximately 20 to 50 mm on human forehead measurements. The regression of phi is clearly reduced in measurements where light leakage occurs, mainly due to insufficient contact between the source optode and tissue. Utilizing this observation, a novel criterion to detect light leakage is developed. The criterion is applied to study the reliability of hemodynamic responses measured on the human forehead when breathing carbon dioxide-enriched air and during hyperventilation. The contrast of the signals is significantly lower in measurements which were adversely affected by light leakage. Furthermore, such unreliable signals at SDDs >or= 50 mm correlate significantly (for [HbO2] p < 0.01 and for [HbR] p < 0.001) better with the signals measured at SDDs < 20 mm. Using this method, poor contact between the source optode and tissue can be detected and corrected before the actual measurement, which enables us to avoid the acquisition of low contrast cortical signals.