Purpose: Several studies have been carried out, and there is no classification for proximal humeral fractures (PHF) exempted from variability in interpretation and with questioned reliability. In the present study, we investigated the ‘absolute diagnostic reliability’ of the most currently used classifications for PHFs on a single anterior-posterior X-ray shoulder image. Methods: Six orthopaedic surgeons, with varying levels of experience in shoulder pathology, evaluated radiographs from 30 proximal humeral fractures, according to the ‘absolute reliability’ criteria. Each of the observers rated each fracture according to Neer, Müller/AO and Codman-Hertel’s classification systems. Results: The overall inter-observer agreement (κ) has been 0.297 (CI95% 0.280 to 0.314) for the Neer’s classification system, 0.206 (CI95% 0.193 to 0.218) for the Müller/AO classification system, and 0.315 (CI95% 0.334 to 0.368) for the Codman-Hertel classification system. We found loss of agreement in Neer’s classification as the study progressed, low agreement in the AO classification, and stable values in the different evaluations with the best degree of agreement for Codman-Hertel classification, with a moderate agreement in the second evaluation among the six evaluators. Conclusion: The Neer, AO, and Hertel-Codman classification systems for PHF with a single radiographic projection have a difficult interpretation for orthopaedic surgeons of varying levels of experience, and therefore substantial agreements are not obtained.