Although the electronic stability control devices have reduced the number of dynamic rollover accidents, it still occupies non-negligible portion of the traffic accidents with fatality and severe injuries. The principal body region of fatal or severe injury in dynamic rollover is cervical spine, while there have been no recognized injury criteria of cervical spinal injuries using existing ATDs for such a loading condition. In this study, the authors tried to establish the method to evaluate cervical spinal injury of the car passengers in dynamic rollover using a human FE model. The human FE model that the authors had developed to be capable of predicting whole body kinematics and the injuries on thorax, lumbar spine, and lower extremities of car occupant in frontal and side impact was adopted as a baseline model. Since the cervical spine part of the model had been constructed by jointed rigid bodies, it could not be used to predict injury level under loading. Therefore, the model was modified to be capable of injury prediction. First, each vertebral body of the cervical spine was modified to deformable, and the deformable intervertebral disk (IVD) was inserted between each pair of vertebral bodies. Second, each isolated vertebra or IVD models were exposed to static compression in the same conditions as the experiments from the literature to find the critical stress corresponding to the fracture or rupture in the experiments. Next, the kinematics and these critical stress values were validated against the whole body inverted drop tests from the literature. Finally, the critical stress values were examined to be available in several different angles of impact in two series of head-neck drop tests from the literature. In the whole body inverted drop, the kinematics of the cervical spine was well replicated by the model and the critical stress values could well divided the impact velocities with or without injuries. In the head-neck drop with different angles of impact, the model could well predict injurious or non-injurious conditions of the tests. In addition, existing anthropomorphic test devices (ATDs) were examined if their neck structures and corresponding injury criteria could be used for evaluating cervical spinal injuries in rollover compared with the human model. It was found that there were large differences between the predicted injury by the modified human model and those by ATDs’ output based on the injury assessment reference value (IARV). As a result, the human FE occupant model modified to have deformable vertebral bodies and IVDs instead of jointed rigid bodies appeared to be capable of predicting cervical spinal injury in dynamic rollover. On the other hand, it could be mentioned that further investigation on ATD neck structure and/or injury criterion is necessary to establish a physical evaluation method for occupant protection in dynamic rollover.