Study of the relationship between hemodynamics and atherogenesis requires accurate three-dimensional descriptions of in vivo arterial geometries. Common methods for obtaining such geometries include in vivo medical imaging and postmortem preparations (vessel casts, pressure-fixed vessels). We sought to determine the relative accuracy of these methods. The aorto-iliac (A/I) region of six rabbits was imaged in vivo using contrast-enhanced magnetic resonance imaging (MRI). After sacrifice, the geometry of the A/I region was preserved via vascular casts in four animals, and ex situ pressure fixation (while preserving dimensions) in the remaining two animals. The MR images and postmortem preparations were used to build computer representations of the A/I bifurcations, which were then used as input for computational blood flow analyses. Substantial differences were seen between MRI-based models and postmortem preparations. Bifurcation angles were consistently larger in postmortem specimens, and vessel dimensions were consistently smaller in pressure-fixed specimens. In vivo MRI-based models underpredicted aortic dimensions immediately proximal to the bifurcation, causing appreciable variation in the aorto-iliac parent/child area ratio. This had an important effect on wall shear stress and separation patterns on the "hips" of the bifurcation, with mean wall shear stress differences ranging from 15% to 35%, depending on the model. The above results, as well as consideration of known and probable sources of error, suggests that in vivo MRI best replicates overall vessel geometry (vessel paths and bifurcation angle). However, vascular casting seems to better capture detailed vessel cross-sectional dimensions and shape. It is important to accurately characterize the local aorto-iliac area ratio when studying in vivo bifurcation hemodynamics.