Flow inside a scaled model of nasal cavity was measured by tomographic PIV. The model was constructed with transparent silicon. A refractive index of working fluid was matched to the model index by mixing glycerol and water. Four cameras and double pulse laser system were used for tomographic PIV. To obtain high SNR, red fluorescence particles and longpass glass filters were used. A complex model shape was estimated by accumulating three-dimensional particle positions obtained by least square based triangulation. Morphological operations such as opening and closing were used to improve the estimation quality. The geometry information was used to distinguish the fluid and solid regions in the tomographic reconstruction procedure. The model estimation and tomographic reconstruction procedures were evaluated by synthetic image test. The synthetic images were generated by placing a computer model (stereolithography file from CT data) to a virtual 3D coordinates and by seeding particles in the flow region randomly. Perspective transformation matrix for synthetic image generation was adopted from the experimental camera matrix. The synthetic image test shows that the reconstruction quality for a complex 3D geometry was improved by a masking technique. Finally, flow velocity field was calculated from 3D cross-correlation of reconstructed voxel intensities.