Near infrared spectrophotometric (NIRS) algorithms to determine the tissue oxygen saturation (TOI) assume a semi-infinite, homogenous tissue geometry. At the head, the clear cerebrospinal fluid (CSF) layer may violate this assumption. The aim was to estimate the error in the TOI values caused by the CSF layer in vitro and to confirm the results in vivo. The liquid phantom mimicking the neonatal head, consisted of a spherical shell of silicone filled with a liquid solution (1% Intralipid, 60 mumol/l haemoglobin, yeast) and a clear layer imitating CSF. The solution was oxygenated and deoxygenated, while measuring its TOI and pO2. Without clear layer the mean TOI was 90.9 +/- 0.5% at pO2 > 18 kPa and decreased to 26.0 +/- 1.3% at pO2 = 0 kPa. With a clear layer the TOI increased from 27.8 +/- 0.8% at pO2 > 18 kPa to 68.0 +/- 0.8% at pO2 = 0 kPa. The clear layer caused a large error in the TOI. In ten mechanically ventilated infants (postnatal age 0.03 to 8 months) the TOI (at the head) and arterial oxygen saturation (SaO2) were measured while the inspired oxygen fraction was altered. The TOI was always positively correlated with the SaO2 (mean slope linear regression = 0.89, r2 = 0.62). Thus an adverse effect of the CSF layer on TOI measurements can be excluded for infants. The CSF layer is not modelled correctly in the phantom.