Proximal humeral fractures can restrict daily activities and, therefore, deserve efficient diagnoses that minimize complications and sequels. For good diagnosis and treatment, patient characteristics, variability in the forms of the fractures presented, and the technical difficulties in achieving fair results with surgical treatment should all be taken into account. Current classification systems for these fractures are based on anatomical and pathological principles, and not on systematic image reading. These fractures can appear in many different forms, with many characteristics that must be identified. However, many current classification systems lack good reliability, both inter-observer and intra-observer for different image types. A new approach to image reading, following a well-designed set and sequence of variables to check, is needed. We previously reported such an image reading system. In the present study, we report a classification system based on this image reading system. Here we define 21 fracture characteristics and apply them along with classical Codman approaches to classify fractures. We base this novel classification system for classifying proximal humeral fractures on a review of scientific literature and improvements to our image reading protocol. Patient status, fracture characteristics and surgeon circumstances have been important issues in developing this system.