To assess the reliability of posture across and within subjects, specifically the repeatability of spinal angles determined by digitization of images in the anterior, posterior, and sagittal views. A repeated measure design was used in which subjects were required to attend 3 sessions, each consisting of 3 trials. Photographs of the anterior, posterior, and lateral views of normal, relaxed upright standing were taken during each trial. Landmarks were digitized and cervical, thoracic, and lumbar angles were calculated with respect to a vertical reference line. Fourteen healthy and active subjects (7 male subjects and 7 female subjects) were recruited from a university student population. All had been free of low back pain during the previous 6 months. When comparing mean angles, no significant differences were detected for any angle in any view. However, large variability within subjects was observed, likely leading to the lack of significance found with respect to the main factors in the analysis of variance (ANOVA). Large coefficients of variance (CVs) reflect the substantial intrasubject variability, as well as poor to moderate agreement indicated by intraclass correlation coefficients (ICCs). There were no apparent trends indicating that gender affected repeatability of posture. The poor repeatability of postures documented using the studied method brings into question the validity of this postural analysis approach for either diagnostic use or tracking changes in response to treatment. Users of such postural analysis tools should interpret postural deviations from a vertical reference with caution, as there are many inherent factors that can contribute to the variability of these measured postures.