BACKGROUND Evidence suggests a relationship between glenoid retroversion and posterior instability, but no literature exists comparing glenoid version referencing the scapular body versus the endosteal vault. This study evaluated glenoid version and its relationship to unidirectional instability and labral tears. METHODS Glenoid version in patients with unidirectional instability or labral tears was measured with magnetic resonance imaging by either the Friedman method or the Poon and Ting method. Analyses of variance followed by independent t tests were used to compare 3 groups: anterior instability or labral tears (anterior pathology group, n = 33); posterior instability or labral tears (posterior pathology group, n = 34); and stable controls (n = 30). The referencing error for 2-dimensional axial images was evaluated for variance by imaging facility. Interobserver and intraobserver reliability scores were calculated. RESULTS With the Friedman method, the posterior pathology group (-9°) was more retroverted than the control group (-4°) (P = .0005) and the anterior pathology group (-5°) (P = .0104) but there was no difference between the control group and anterior pathology group (P = .38). The referencing error in the sagittal plane averaged 23° and varied by facility (P = .0365). The coronal-plane error averaged 1° and did not vary by facility (P = .7180). Intraclass correlation coefficient scores showed good to excellent intrarater and inter-rater reliability. CONCLUSIONS The posterior pathology group had 5° more retroversion than controls using the Friedman method. Glenoid version using the Poon and Ting method or the Friedman method did not predict anterior instability or labral tears. Axial magnetic resonance images were constructed with a referencing error in the sagittal plane that varied by magnetic resonance imaging facility and has implications for improving 2-dimensional axial imaging protocols.