Recent studies have demonstrated inconsistencies in the use of certain images for classifying proximal humerus fractures. Our purpose was to determine whether three-dimensional computed tomography or the level of expertise of the observers would improve the reliability and reproducibility of identifying specific anatomic fragments in proximal humerus fractures. Two groups of observers, nonexperts and experts in shoulder surgery, were asked to review the radiographs and three-dimensional computed tomography scans of 12 patients with proximal humerus fractures. Observers were asked to identify displaced fracture fragments, dislocation, and articular surface fractures. Both groups of observers displayed suboptimal reliability for the identification of displaced fracture fragments. The addition of three-dimensional computed tomography scans did not improve the reliability or reproducibility. Poor agreement for the purpose of classification seems to occur at the most fundamental level, the pathoanatomic description of the fracture. Inconsistencies may have been due to imprecise identification and measurement of individual fracture fragments, differing interpretations of the pathoanatomy, or both.