Accurate analysis of osseous glenoid morphology is important in treating glenohumeral arthritis and instability. Two-dimensional computed tomography scans are used to evaluate glenoid alignment. Accuracy of this method is dependent on the angle of axial reconstruction in relation to the position of the scapula. The purpose of this study was to investigate the effect of scapular rotation in the coronal and sagittal planes on glenoid version as measured on two-dimensional images. Computer-generated three-dimensional models of scapulae from computed tomography scans of thirty-six shoulders in whole-body cadavers were generated. The anatomic geometry of these models had been previously validated. The position of the scapulae relative to the gantry was determined. The three-dimensional models were rotated in 1 degree increments in the coronal and sagittal planes. Glenoid version was measured on two-dimensional images for each of the rotation increments. Version variability at each rotation increment was calculated. The anatomic glenoid version (independent of the resting position of the scapula) was an average (and standard deviation) of 2.0 degrees +/- 3.8 degrees of retroversion. The average difference between anatomic glenoid version and clinical glenoid version (depending on the position of the scapula on the original computed tomography axial images) was 6.9 degrees +/- 5.6 degrees (range, 0.1 degrees to 22.5 degrees). Version variability with coronal or sagittal rotation was significant for all degrees of rotation (p < 0.0001). Scapular abduction had the greatest effect on version variation and resulted in 0.42 degrees of relative anteversion for every 1 degree of abduction in the coronal plane. In the sagittal plane, internal rotation resulted in relative anteversion. Any malalignment of > or = 1 degree of the scapula in the coronal or sagittal plane will create inaccuracies in measuring glenoid version. The plane of axial reconstruction should be aligned with the scapula when two-dimensional computed tomography images are used to measure glenoid version. These findings support the use of three-dimensional models to evaluate glenoid version.