The safety of new autonomous vehicles must be evaluated before introducing them into mixed traffic. But before this can be done, safety-critical scenarios must be identified and tested. Using expert knowledge and field data we identify ten safety critical scenarios to simulate and therefore evaluate the autonomous vehicle performance compared to conventional, human-driven vehicles performance. This selection was made balancing the criticality and frequency of a given situation. To describe these scenarios, we create an ontology that considers the difference between autonomous vehicles and human drivers and provides an exhaustive description of the vehicle surrounding. In this paper, we propose a generic framework and ontology usable for safety-critical scenarios description and generation.