A detailed MCNP neutronics model of the building housing the TCV tokamak has been developed. Predicted dose rates have been compared to neutron dose measurements at 12 locations in the building obtained during neutral beam injection into TCV plasmas. This work was motivated both by the necessity to understand neutron transport in a complex environment and to provide an adequate reference model of the building to which additional shielding can be added. The additional shielding is required because of an expected 5-10 fold increase of neutron doses by the addition of a second neutral beam source. The neutron source was estimated using the TRANSP code. Starting from a simple model, which left out or simplified many important structures, the model was iteratively refined. The discrepancies between the predicted and measured doses decreased significantly, down to less than a factor 2 for most of the locations calculated by the most detailed model. A factor 2 is believed to be within the combined errors resulting from uncertainties in the neutron source model, the detector calibration and the uncertainties stemming from the model itself.