Bone breakage is one of the most common features in the archaeological record. Fractures occur at different times and are classified as fresh or dry depending on the presence or absence of collagen in the bone. In the study of human remains, the timing of the occurrence of a fracture is of crucial importance as it can sometimes be linked to the cause of death. Types of skull breakage can be classified based on when they occurred, though not all fractures correspond to the expected features. This variability is added to the challenge of working with bones covered in consolidant, which obstructs the bone surface and hinders taphonomic analysis. This is the case of the Txispiri calotte, which was categorized as a skull cup in the early 20th century, though this classification was later rejected in the 1990s. In this study, we used statistics and machine learning (ML) to test the breakage characteristics of one set of skull fragments with fresh fractures, another set with dry fractures, and the Txispiri calotte. For this purpose, we considered the fracture type, trajectory, angles, cortical delamination and texture of each of the individual fractures. Our results show that the 13 fractures of the Txispiri calotte correspond to dry breakage and bear no relation to artificially produced skull cups. This study shows the potential of ML algorithms to classify fresh and dry fractures within the same specimen, a method that can be applied to other assemblages with similar characteristics.