The object of the present study was to compare stereological estimates of brain volumes obtained in vivo by magnetic resonance imaging (MRI) to corresponding volumes from physical sections in vitro. Brains of ten domestic pigs were imaged using a 3-T scanner. The volumes of different brain compartments were obtained from MR images by two observers and from physical sections using the Cavalieri estimator in combination with point counting. Paired t tests revealed no significant differences between the two methods for any of the five compartments considered, except for the basal gray compartment. However, although intraobserver difference of MRI estimates was acceptable, the interobserver difference was not. A statistical highly significant difference of 11-41% was observed between observers for volume estimates of all compartments considered. The study demonstrates that quantitative MRI is susceptible to observer dependent interpretation of images