This study tests a multiwavelength frequency-domain near-infrared oximeter (fdNIRS) in an in vitro model of the human brain. The model is a solid plastic structure containing a vascular network perfused with blood in which haemoglobin oxygen saturation (SO2) was measured by co-oximetry, providing a standard for comparison. Plastic shells of varying thickness (0.5-2 cm), with a vascular network of their own and encircling the brain model, were also added to simulate extracranial tissues of the infant, child and adult. The fdNIRS oximeter utilizes frequency-domain technology to monitor phaseshifts at 754 nm, 785 nm and 816 nm relative to a 780 nm reference to derive SO2 through photon transport and Beer-Lambert equations. We found a linear relationship between fdNIRS SO2 and co-oximetry SO2 with excellent correlation (r2 > or = 0.95) that fitted the line of identity in all experiments (n = 7). The bias of fdNIRS oximetry was -2% and the precision was 6%. Blood temperature and fdNIRS source-detector distance did not affect fdNIRS oximetry. Low haemoglobin concentration (6 g dl(-1)) altered the fdNIRS versus co-oximetry line slope and intercept, producing a 15% error at the extremes of SO2. The infant- and child-like shells overlying the brain model did not alter fdNIRS oximetry, whereas the adult-like shell yielded an error as high as 32%. In conclusion, fdNIRS accurately measures SO2 in an in vitro brain model, although low haemoglobin concentration and extracranial tissue of adult thickness influence accuracy.